├── .idea ├── .gitignore ├── Optimal-Controller-for-Multiagent-USVs.iml ├── inspectionProfiles │ └── profiles_settings.xml ├── misc.xml ├── modules.xml └── vcs.xml ├── LICENSE.md ├── README.md ├── documents └── prove.pdf └── problem1 ├── adp_drl_nn.py ├── agent_def.py ├── main.py ├── model_def.py ├── rbf.py ├── rbf_network.py ├── reg.py ├── test_unit1.py ├── test_unit2.py ├── test_unit3.py ├── test_unit4.py └── usvs_control.py /.idea/.gitignore: -------------------------------------------------------------------------------- 1 | # Default ignored files 2 | /shelf/ 3 | /workspace.xml 4 | -------------------------------------------------------------------------------- /.idea/Optimal-Controller-for-Multiagent-USVs.iml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | -------------------------------------------------------------------------------- /.idea/inspectionProfiles/profiles_settings.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 6 | -------------------------------------------------------------------------------- /.idea/misc.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | -------------------------------------------------------------------------------- /.idea/modules.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | -------------------------------------------------------------------------------- /.idea/vcs.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | -------------------------------------------------------------------------------- /LICENSE.md: -------------------------------------------------------------------------------- 1 | GNU GENERAL PUBLIC LICENSE 2 | Version 3, 29 June 2007 3 | 4 | Copyright (C) 2007 Free Software Foundation, Inc. 5 | Everyone is permitted to copy and distribute verbatim copies 6 | of this license document, but changing it is not allowed. 7 | 8 | Preamble 9 | 10 | The GNU General Public License is a free, copyleft license for 11 | software and other kinds of works. 12 | 13 | The licenses for most software and other practical works are designed 14 | to take away your freedom to share and change the works. By contrast, 15 | the GNU General Public License is intended to guarantee your freedom to 16 | share and change all versions of a program--to make sure it remains free 17 | software for all its users. We, the Free Software Foundation, use the 18 | GNU General Public License for most of our software; it applies also to 19 | any other work released this way by its authors. You can apply it to 20 | your programs, too. 21 | 22 | When we speak of free software, we are referring to freedom, not 23 | price. Our General Public Licenses are designed to make sure that you 24 | have the freedom to distribute copies of free software (and charge for 25 | them if you wish), that you receive source code or can get it if you 26 | want it, that you can change the software or use pieces of it in new 27 | free programs, and that you know you can do these things. 28 | 29 | To protect your rights, we need to prevent others from denying you 30 | these rights or asking you to surrender the rights. Therefore, you have 31 | certain responsibilities if you distribute copies of the software, or if 32 | you modify it: responsibilities to respect the freedom of others. 33 | 34 | For example, if you distribute copies of such a program, whether 35 | gratis or for a fee, you must pass on to the recipients the same 36 | freedoms that you received. You must make sure that they, too, receive 37 | or can get the source code. And you must show them these terms so they 38 | know their rights. 39 | 40 | Developers that use the GNU GPL protect your rights with two steps: 41 | (1) assert copyright on the software, and (2) offer you this License 42 | giving you legal permission to copy, distribute and/or modify it. 43 | 44 | For the developers' and authors' protection, the GPL clearly explains 45 | that there is no warranty for this free software. For both users' and 46 | authors' sake, the GPL requires that modified versions be marked as 47 | changed, so that their problems will not be attributed erroneously to 48 | authors of previous versions. 49 | 50 | Some devices are designed to deny users access to install or run 51 | modified versions of the software inside them, although the manufacturer 52 | can do so. This is fundamentally incompatible with the aim of 53 | protecting users' freedom to change the software. The systematic 54 | pattern of such abuse occurs in the area of products for individuals to 55 | use, which is precisely where it is most unacceptable. Therefore, we 56 | have designed this version of the GPL to prohibit the practice for those 57 | products. If such problems arise substantially in other domains, we 58 | stand ready to extend this provision to those domains in future versions 59 | of the GPL, as needed to protect the freedom of users. 60 | 61 | Finally, every program is threatened constantly by software patents. 62 | States should not allow patents to restrict development and use of 63 | software on general-purpose computers, but in those that do, we wish to 64 | avoid the special danger that patents applied to a free program could 65 | make it effectively proprietary. To prevent this, the GPL assures that 66 | patents cannot be used to render the program non-free. 67 | 68 | The precise terms and conditions for copying, distribution and 69 | modification follow. 70 | 71 | TERMS AND CONDITIONS 72 | 73 | 0. Definitions. 74 | 75 | "This License" refers to version 3 of the GNU General Public License. 76 | 77 | "Copyright" also means copyright-like laws that apply to other kinds of 78 | works, such as semiconductor masks. 79 | 80 | "The Program" refers to any copyrightable work licensed under this 81 | License. Each licensee is addressed as "you". "Licensees" and 82 | "recipients" may be individuals or organizations. 83 | 84 | To "modify" a work means to copy from or adapt all or part of the work 85 | in a fashion requiring copyright permission, other than the making of an 86 | exact copy. The resulting work is called a "modified version" of the 87 | earlier work or a work "based on" the earlier work. 88 | 89 | A "covered work" means either the unmodified Program or a work based 90 | on the Program. 91 | 92 | To "propagate" a work means to do anything with it that, without 93 | permission, would make you directly or secondarily liable for 94 | infringement under applicable copyright law, except executing it on a 95 | computer or modifying a private copy. Propagation includes copying, 96 | distribution (with or without modification), making available to the 97 | public, and in some countries other activities as well. 98 | 99 | To "convey" a work means any kind of propagation that enables other 100 | parties to make or receive copies. Mere interaction with a user through 101 | a computer network, with no transfer of a copy, is not conveying. 102 | 103 | An interactive user interface displays "Appropriate Legal Notices" 104 | to the extent that it includes a convenient and prominently visible 105 | feature that (1) displays an appropriate copyright notice, and (2) 106 | tells the user that there is no warranty for the work (except to the 107 | extent that warranties are provided), that licensees may convey the 108 | work under this License, and how to view a copy of this License. If 109 | the interface presents a list of user commands or options, such as a 110 | menu, a prominent item in the list meets this criterion. 111 | 112 | 1. Source Code. 113 | 114 | The "source code" for a work means the preferred form of the work 115 | for making modifications to it. "Object code" means any non-source 116 | form of a work. 117 | 118 | A "Standard Interface" means an interface that either is an official 119 | standard defined by a recognized standards body, or, in the case of 120 | interfaces specified for a particular programming language, one that 121 | is widely used among developers working in that language. 122 | 123 | The "System Libraries" of an executable work include anything, other 124 | than the work as a whole, that (a) is included in the normal form of 125 | packaging a Major Component, but which is not part of that Major 126 | Component, and (b) serves only to enable use of the work with that 127 | Major Component, or to implement a Standard Interface for which an 128 | implementation is available to the public in source code form. A 129 | "Major Component", in this context, means a major essential component 130 | (kernel, window system, and so on) of the specific operating system 131 | (if any) on which the executable work runs, or a compiler used to 132 | produce the work, or an object code interpreter used to run it. 133 | 134 | The "Corresponding Source" for a work in object code form means all 135 | the source code needed to generate, install, and (for an executable 136 | work) run the object code and to modify the work, including scripts to 137 | control those activities. However, it does not include the work's 138 | System Libraries, or general-purpose tools or generally available free 139 | programs which are used unmodified in performing those activities but 140 | which are not part of the work. For example, Corresponding Source 141 | includes interface definition files associated with source files for 142 | the work, and the source code for shared libraries and dynamically 143 | linked subprograms that the work is specifically designed to require, 144 | such as by intimate data communication or control flow between those 145 | subprograms and other parts of the work. 146 | 147 | The Corresponding Source need not include anything that users 148 | can regenerate automatically from other parts of the Corresponding 149 | Source. 150 | 151 | The Corresponding Source for a work in source code form is that 152 | same work. 153 | 154 | 2. Basic Permissions. 155 | 156 | All rights granted under this License are granted for the term of 157 | copyright on the Program, and are irrevocable provided the stated 158 | conditions are met. This License explicitly affirms your unlimited 159 | permission to run the unmodified Program. The output from running a 160 | covered work is covered by this License only if the output, given its 161 | content, constitutes a covered work. This License acknowledges your 162 | rights of fair use or other equivalent, as provided by copyright law. 163 | 164 | You may make, run and propagate covered works that you do not 165 | convey, without conditions so long as your license otherwise remains 166 | in force. You may convey covered works to others for the sole purpose 167 | of having them make modifications exclusively for you, or provide you 168 | with facilities for running those works, provided that you comply with 169 | the terms of this License in conveying all material for which you do 170 | not control copyright. Those thus making or running the covered works 171 | for you must do so exclusively on your behalf, under your direction 172 | and control, on terms that prohibit them from making any copies of 173 | your copyrighted material outside their relationship with you. 174 | 175 | Conveying under any other circumstances is permitted solely under 176 | the conditions stated below. Sublicensing is not allowed; section 10 177 | makes it unnecessary. 178 | 179 | 3. Protecting Users' Legal Rights From Anti-Circumvention Law. 180 | 181 | No covered work shall be deemed part of an effective technological 182 | measure under any applicable law fulfilling obligations under article 183 | 11 of the WIPO copyright treaty adopted on 20 December 1996, or 184 | similar laws prohibiting or restricting circumvention of such 185 | measures. 186 | 187 | When you convey a covered work, you waive any legal power to forbid 188 | circumvention of technological measures to the extent such circumvention 189 | is effected by exercising rights under this License with respect to 190 | the covered work, and you disclaim any intention to limit operation or 191 | modification of the work as a means of enforcing, against the work's 192 | users, your or third parties' legal rights to forbid circumvention of 193 | technological measures. 194 | 195 | 4. Conveying Verbatim Copies. 196 | 197 | You may convey verbatim copies of the Program's source code as you 198 | receive it, in any medium, provided that you conspicuously and 199 | appropriately publish on each copy an appropriate copyright notice; 200 | keep intact all notices stating that this License and any 201 | non-permissive terms added in accord with section 7 apply to the code; 202 | keep intact all notices of the absence of any warranty; and give all 203 | recipients a copy of this License along with the Program. 204 | 205 | You may charge any price or no price for each copy that you convey, 206 | and you may offer support or warranty protection for a fee. 207 | 208 | 5. Conveying Modified Source Versions. 209 | 210 | You may convey a work based on the Program, or the modifications to 211 | produce it from the Program, in the form of source code under the 212 | terms of section 4, provided that you also meet all of these conditions: 213 | 214 | a) The work must carry prominent notices stating that you modified 215 | it, and giving a relevant date. 216 | 217 | b) The work must carry prominent notices stating that it is 218 | released under this License and any conditions added under section 219 | 7. This requirement modifies the requirement in section 4 to 220 | "keep intact all notices". 221 | 222 | c) You must license the entire work, as a whole, under this 223 | License to anyone who comes into possession of a copy. This 224 | License will therefore apply, along with any applicable section 7 225 | additional terms, to the whole of the work, and all its parts, 226 | regardless of how they are packaged. This License gives no 227 | permission to license the work in any other way, but it does not 228 | invalidate such permission if you have separately received it. 229 | 230 | d) If the work has interactive user interfaces, each must display 231 | Appropriate Legal Notices; however, if the Program has interactive 232 | interfaces that do not display Appropriate Legal Notices, your 233 | work need not make them do so. 234 | 235 | A compilation of a covered work with other separate and independent 236 | works, which are not by their nature extensions of the covered work, 237 | and which are not combined with it such as to form a larger program, 238 | in or on a volume of a storage or distribution medium, is called an 239 | "aggregate" if the compilation and its resulting copyright are not 240 | used to limit the access or legal rights of the compilation's users 241 | beyond what the individual works permit. Inclusion of a covered work 242 | in an aggregate does not cause this License to apply to the other 243 | parts of the aggregate. 244 | 245 | 6. Conveying Non-Source Forms. 246 | 247 | You may convey a covered work in object code form under the terms 248 | of sections 4 and 5, provided that you also convey the 249 | machine-readable Corresponding Source under the terms of this License, 250 | in one of these ways: 251 | 252 | a) Convey the object code in, or embodied in, a physical product 253 | (including a physical distribution medium), accompanied by the 254 | Corresponding Source fixed on a durable physical medium 255 | customarily used for software interchange. 256 | 257 | b) Convey the object code in, or embodied in, a physical product 258 | (including a physical distribution medium), accompanied by a 259 | written offer, valid for at least three years and valid for as 260 | long as you offer spare parts or customer support for that product 261 | model, to give anyone who possesses the object code either (1) a 262 | copy of the Corresponding Source for all the software in the 263 | product that is covered by this License, on a durable physical 264 | medium customarily used for software interchange, for a price no 265 | more than your reasonable cost of physically performing this 266 | conveying of source, or (2) access to copy the 267 | Corresponding Source from a network server at no charge. 268 | 269 | c) Convey individual copies of the object code with a copy of the 270 | written offer to provide the Corresponding Source. This 271 | alternative is allowed only occasionally and noncommercially, and 272 | only if you received the object code with such an offer, in accord 273 | with subsection 6b. 274 | 275 | d) Convey the object code by offering access from a designated 276 | place (gratis or for a charge), and offer equivalent access to the 277 | Corresponding Source in the same way through the same place at no 278 | further charge. You need not require recipients to copy the 279 | Corresponding Source along with the object code. If the place to 280 | copy the object code is a network server, the Corresponding Source 281 | may be on a different server (operated by you or a third party) 282 | that supports equivalent copying facilities, provided you maintain 283 | clear directions next to the object code saying where to find the 284 | Corresponding Source. Regardless of what server hosts the 285 | Corresponding Source, you remain obligated to ensure that it is 286 | available for as long as needed to satisfy these requirements. 287 | 288 | e) Convey the object code using peer-to-peer transmission, provided 289 | you inform other peers where the object code and Corresponding 290 | Source of the work are being offered to the general public at no 291 | charge under subsection 6d. 292 | 293 | A separable portion of the object code, whose source code is excluded 294 | from the Corresponding Source as a System Library, need not be 295 | included in conveying the object code work. 296 | 297 | A "User Product" is either (1) a "consumer product", which means any 298 | tangible personal property which is normally used for personal, family, 299 | or household purposes, or (2) anything designed or sold for incorporation 300 | into a dwelling. In determining whether a product is a consumer product, 301 | doubtful cases shall be resolved in favor of coverage. For a particular 302 | product received by a particular user, "normally used" refers to a 303 | typical or common use of that class of product, regardless of the status 304 | of the particular user or of the way in which the particular user 305 | actually uses, or expects or is expected to use, the product. A product 306 | is a consumer product regardless of whether the product has substantial 307 | commercial, industrial or non-consumer uses, unless such uses represent 308 | the only significant mode of use of the product. 309 | 310 | "Installation Information" for a User Product means any methods, 311 | procedures, authorization keys, or other information required to install 312 | and execute modified versions of a covered work in that User Product from 313 | a modified version of its Corresponding Source. The information must 314 | suffice to ensure that the continued functioning of the modified object 315 | code is in no case prevented or interfered with solely because 316 | modification has been made. 317 | 318 | If you convey an object code work under this section in, or with, or 319 | specifically for use in, a User Product, and the conveying occurs as 320 | part of a transaction in which the right of possession and use of the 321 | User Product is transferred to the recipient in perpetuity or for a 322 | fixed term (regardless of how the transaction is characterized), the 323 | Corresponding Source conveyed under this section must be accompanied 324 | by the Installation Information. But this requirement does not apply 325 | if neither you nor any third party retains the ability to install 326 | modified object code on the User Product (for example, the work has 327 | been installed in ROM). 328 | 329 | The requirement to provide Installation Information does not include a 330 | requirement to continue to provide support service, warranty, or updates 331 | for a work that has been modified or installed by the recipient, or for 332 | the User Product in which it has been modified or installed. Access to a 333 | network may be denied when the modification itself materially and 334 | adversely affects the operation of the network or violates the rules and 335 | protocols for communication across the network. 336 | 337 | Corresponding Source conveyed, and Installation Information provided, 338 | in accord with this section must be in a format that is publicly 339 | documented (and with an implementation available to the public in 340 | source code form), and must require no special password or key for 341 | unpacking, reading or copying. 342 | 343 | 7. Additional Terms. 344 | 345 | "Additional permissions" are terms that supplement the terms of this 346 | License by making exceptions from one or more of its conditions. 347 | Additional permissions that are applicable to the entire Program shall 348 | be treated as though they were included in this License, to the extent 349 | that they are valid under applicable law. If additional permissions 350 | apply only to part of the Program, that part may be used separately 351 | under those permissions, but the entire Program remains governed by 352 | this License without regard to the additional permissions. 353 | 354 | When you convey a copy of a covered work, you may at your option 355 | remove any additional permissions from that copy, or from any part of 356 | it. (Additional permissions may be written to require their own 357 | removal in certain cases when you modify the work.) You may place 358 | additional permissions on material, added by you to a covered work, 359 | for which you have or can give appropriate copyright permission. 360 | 361 | Notwithstanding any other provision of this License, for material you 362 | add to a covered work, you may (if authorized by the copyright holders of 363 | that material) supplement the terms of this License with terms: 364 | 365 | a) Disclaiming warranty or limiting liability differently from the 366 | terms of sections 15 and 16 of this License; or 367 | 368 | b) Requiring preservation of specified reasonable legal notices or 369 | author attributions in that material or in the Appropriate Legal 370 | Notices displayed by works containing it; or 371 | 372 | c) Prohibiting misrepresentation of the origin of that material, or 373 | requiring that modified versions of such material be marked in 374 | reasonable ways as different from the original version; or 375 | 376 | d) Limiting the use for publicity purposes of names of licensors or 377 | authors of the material; or 378 | 379 | e) Declining to grant rights under trademark law for use of some 380 | trade names, trademarks, or service marks; or 381 | 382 | f) Requiring indemnification of licensors and authors of that 383 | material by anyone who conveys the material (or modified versions of 384 | it) with contractual assumptions of liability to the recipient, for 385 | any liability that these contractual assumptions directly impose on 386 | those licensors and authors. 387 | 388 | All other non-permissive additional terms are considered "further 389 | restrictions" within the meaning of section 10. If the Program as you 390 | received it, or any part of it, contains a notice stating that it is 391 | governed by this License along with a term that is a further 392 | restriction, you may remove that term. If a license document contains 393 | a further restriction but permits relicensing or conveying under this 394 | License, you may add to a covered work material governed by the terms 395 | of that license document, provided that the further restriction does 396 | not survive such relicensing or conveying. 397 | 398 | If you add terms to a covered work in accord with this section, you 399 | must place, in the relevant source files, a statement of the 400 | additional terms that apply to those files, or a notice indicating 401 | where to find the applicable terms. 402 | 403 | Additional terms, permissive or non-permissive, may be stated in the 404 | form of a separately written license, or stated as exceptions; 405 | the above requirements apply either way. 406 | 407 | 8. Termination. 408 | 409 | You may not propagate or modify a covered work except as expressly 410 | provided under this License. Any attempt otherwise to propagate or 411 | modify it is void, and will automatically terminate your rights under 412 | this License (including any patent licenses granted under the third 413 | paragraph of section 11). 414 | 415 | However, if you cease all violation of this License, then your 416 | license from a particular copyright holder is reinstated (a) 417 | provisionally, unless and until the copyright holder explicitly and 418 | finally terminates your license, and (b) permanently, if the copyright 419 | holder fails to notify you of the violation by some reasonable means 420 | prior to 60 days after the cessation. 421 | 422 | Moreover, your license from a particular copyright holder is 423 | reinstated permanently if the copyright holder notifies you of the 424 | violation by some reasonable means, this is the first time you have 425 | received notice of violation of this License (for any work) from that 426 | copyright holder, and you cure the violation prior to 30 days after 427 | your receipt of the notice. 428 | 429 | Termination of your rights under this section does not terminate the 430 | licenses of parties who have received copies or rights from you under 431 | this License. If your rights have been terminated and not permanently 432 | reinstated, you do not qualify to receive new licenses for the same 433 | material under section 10. 434 | 435 | 9. Acceptance Not Required for Having Copies. 436 | 437 | You are not required to accept this License in order to receive or 438 | run a copy of the Program. Ancillary propagation of a covered work 439 | occurring solely as a consequence of using peer-to-peer transmission 440 | to receive a copy likewise does not require acceptance. However, 441 | nothing other than this License grants you permission to propagate or 442 | modify any covered work. These actions infringe copyright if you do 443 | not accept this License. Therefore, by modifying or propagating a 444 | covered work, you indicate your acceptance of this License to do so. 445 | 446 | 10. Automatic Licensing of Downstream Recipients. 447 | 448 | Each time you convey a covered work, the recipient automatically 449 | receives a license from the original licensors, to run, modify and 450 | propagate that work, subject to this License. You are not responsible 451 | for enforcing compliance by third parties with this License. 452 | 453 | An "entity transaction" is a transaction transferring control of an 454 | organization, or substantially all assets of one, or subdividing an 455 | organization, or merging organizations. If propagation of a covered 456 | work results from an entity transaction, each party to that 457 | transaction who receives a copy of the work also receives whatever 458 | licenses to the work the party's predecessor in interest had or could 459 | give under the previous paragraph, plus a right to possession of the 460 | Corresponding Source of the work from the predecessor in interest, if 461 | the predecessor has it or can get it with reasonable efforts. 462 | 463 | You may not impose any further restrictions on the exercise of the 464 | rights granted or affirmed under this License. For example, you may 465 | not impose a license fee, royalty, or other charge for exercise of 466 | rights granted under this License, and you may not initiate litigation 467 | (including a cross-claim or counterclaim in a lawsuit) alleging that 468 | any patent claim is infringed by making, using, selling, offering for 469 | sale, or importing the Program or any portion of it. 470 | 471 | 11. Patents. 472 | 473 | A "contributor" is a copyright holder who authorizes use under this 474 | License of the Program or a work on which the Program is based. The 475 | work thus licensed is called the contributor's "contributor version". 476 | 477 | A contributor's "essential patent claims" are all patent claims 478 | owned or controlled by the contributor, whether already acquired or 479 | hereafter acquired, that would be infringed by some manner, permitted 480 | by this License, of making, using, or selling its contributor version, 481 | but do not include claims that would be infringed only as a 482 | consequence of further modification of the contributor version. For 483 | purposes of this definition, "control" includes the right to grant 484 | patent sublicenses in a manner consistent with the requirements of 485 | this License. 486 | 487 | Each contributor grants you a non-exclusive, worldwide, royalty-free 488 | patent license under the contributor's essential patent claims, to 489 | make, use, sell, offer for sale, import and otherwise run, modify and 490 | propagate the contents of its contributor version. 491 | 492 | In the following three paragraphs, a "patent license" is any express 493 | agreement or commitment, however denominated, not to enforce a patent 494 | (such as an express permission to practice a patent or covenant not to 495 | sue for patent infringement). To "grant" such a patent license to a 496 | party means to make such an agreement or commitment not to enforce a 497 | patent against the party. 498 | 499 | If you convey a covered work, knowingly relying on a patent license, 500 | and the Corresponding Source of the work is not available for anyone 501 | to copy, free of charge and under the terms of this License, through a 502 | publicly available network server or other readily accessible means, 503 | then you must either (1) cause the Corresponding Source to be so 504 | available, or (2) arrange to deprive yourself of the benefit of the 505 | patent license for this particular work, or (3) arrange, in a manner 506 | consistent with the requirements of this License, to extend the patent 507 | license to downstream recipients. "Knowingly relying" means you have 508 | actual knowledge that, but for the patent license, your conveying the 509 | covered work in a country, or your recipient's use of the covered work 510 | in a country, would infringe one or more identifiable patents in that 511 | country that you have reason to believe are valid. 512 | 513 | If, pursuant to or in connection with a single transaction or 514 | arrangement, you convey, or propagate by procuring conveyance of, a 515 | covered work, and grant a patent license to some of the parties 516 | receiving the covered work authorizing them to use, propagate, modify 517 | or convey a specific copy of the covered work, then the patent license 518 | you grant is automatically extended to all recipients of the covered 519 | work and works based on it. 520 | 521 | A patent license is "discriminatory" if it does not include within 522 | the scope of its coverage, prohibits the exercise of, or is 523 | conditioned on the non-exercise of one or more of the rights that are 524 | specifically granted under this License. You may not convey a covered 525 | work if you are a party to an arrangement with a third party that is 526 | in the business of distributing software, under which you make payment 527 | to the third party based on the extent of your activity of conveying 528 | the work, and under which the third party grants, to any of the 529 | parties who would receive the covered work from you, a discriminatory 530 | patent license (a) in connection with copies of the covered work 531 | conveyed by you (or copies made from those copies), or (b) primarily 532 | for and in connection with specific products or compilations that 533 | contain the covered work, unless you entered into that arrangement, 534 | or that patent license was granted, prior to 28 March 2007. 535 | 536 | Nothing in this License shall be construed as excluding or limiting 537 | any implied license or other defenses to infringement that may 538 | otherwise be available to you under applicable patent law. 539 | 540 | 12. No Surrender of Others' Freedom. 541 | 542 | If conditions are imposed on you (whether by court order, agreement or 543 | otherwise) that contradict the conditions of this License, they do not 544 | excuse you from the conditions of this License. If you cannot convey a 545 | covered work so as to satisfy simultaneously your obligations under this 546 | License and any other pertinent obligations, then as a consequence you may 547 | not convey it at all. For example, if you agree to terms that obligate you 548 | to collect a royalty for further conveying from those to whom you convey 549 | the Program, the only way you could satisfy both those terms and this 550 | License would be to refrain entirely from conveying the Program. 551 | 552 | 13. Use with the GNU Affero General Public License. 553 | 554 | Notwithstanding any other provision of this License, you have 555 | permission to link or combine any covered work with a work licensed 556 | under version 3 of the GNU Affero General Public License into a single 557 | combined work, and to convey the resulting work. The terms of this 558 | License will continue to apply to the part which is the covered work, 559 | but the special requirements of the GNU Affero General Public License, 560 | section 13, concerning interaction through a network will apply to the 561 | combination as such. 562 | 563 | 14. Revised Versions of this License. 564 | 565 | The Free Software Foundation may publish revised and/or new versions of 566 | the GNU General Public License from time to time. Such new versions will 567 | be similar in spirit to the present version, but may differ in detail to 568 | address new problems or concerns. 569 | 570 | Each version is given a distinguishing version number. If the 571 | Program specifies that a certain numbered version of the GNU General 572 | Public License "or any later version" applies to it, you have the 573 | option of following the terms and conditions either of that numbered 574 | version or of any later version published by the Free Software 575 | Foundation. If the Program does not specify a version number of the 576 | GNU General Public License, you may choose any version ever published 577 | by the Free Software Foundation. 578 | 579 | If the Program specifies that a proxy can decide which future 580 | versions of the GNU General Public License can be used, that proxy's 581 | public statement of acceptance of a version permanently authorizes you 582 | to choose that version for the Program. 583 | 584 | Later license versions may give you additional or different 585 | permissions. However, no additional obligations are imposed on any 586 | author or copyright holder as a result of your choosing to follow a 587 | later version. 588 | 589 | 15. Disclaimer of Warranty. 590 | 591 | THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY 592 | APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT 593 | HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY 594 | OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, 595 | THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 596 | PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM 597 | IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF 598 | ALL NECESSARY SERVICING, REPAIR OR CORRECTION. 599 | 600 | 16. Limitation of Liability. 601 | 602 | IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING 603 | WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS 604 | THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY 605 | GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE 606 | USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF 607 | DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD 608 | PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), 609 | EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF 610 | SUCH DAMAGES. 611 | 612 | 17. Interpretation of Sections 15 and 16. 613 | 614 | If the disclaimer of warranty and limitation of liability provided 615 | above cannot be given local legal effect according to their terms, 616 | reviewing courts shall apply local law that most closely approximates 617 | an absolute waiver of all civil liability in connection with the 618 | Program, unless a warranty or assumption of liability accompanies a 619 | copy of the Program in return for a fee. 620 | 621 | END OF TERMS AND CONDITIONS 622 | 623 | How to Apply These Terms to Your New Programs 624 | 625 | If you develop a new program, and you want it to be of the greatest 626 | possible use to the public, the best way to achieve this is to make it 627 | free software which everyone can redistribute and change under these terms. 628 | 629 | To do so, attach the following notices to the program. It is safest 630 | to attach them to the start of each source file to most effectively 631 | state the exclusion of warranty; and each file should have at least 632 | the "copyright" line and a pointer to where the full notice is found. 633 | 634 | 635 | Copyright (C) 636 | 637 | This program is free software: you can redistribute it and/or modify 638 | it under the terms of the GNU General Public License as published by 639 | the Free Software Foundation, either version 3 of the License, or 640 | (at your option) any later version. 641 | 642 | This program is distributed in the hope that it will be useful, 643 | but WITHOUT ANY WARRANTY; without even the implied warranty of 644 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 645 | GNU General Public License for more details. 646 | 647 | You should have received a copy of the GNU General Public License 648 | along with this program. If not, see . 649 | 650 | Also add information on how to contact you by electronic and paper mail. 651 | 652 | If the program does terminal interaction, make it output a short 653 | notice like this when it starts in an interactive mode: 654 | 655 | Copyright (C) 656 | This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'. 657 | This is free software, and you are welcome to redistribute it 658 | under certain conditions; type `show c' for details. 659 | 660 | The hypothetical commands `show w' and `show c' should show the appropriate 661 | parts of the General Public License. Of course, your program's commands 662 | might be different; for a GUI interface, you would use an "about box". 663 | 664 | You should also get your employer (if you work as a programmer) or school, 665 | if any, to sign a "copyright disclaimer" for the program, if necessary. 666 | For more information on this, and how to apply and follow the GNU GPL, see 667 | . 668 | 669 | The GNU General Public License does not permit incorporating your program 670 | into proprietary programs. If your program is a subroutine library, you 671 | may consider it more useful to permit linking proprietary applications with 672 | the library. If this is what you want to do, use the GNU Lesser General 673 | Public License instead of this License. But first, please read 674 | . 675 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # Optimal-Controller-for-Multiagent-USVs 2 | A Project intended to design a Set of algorithm for Research on Multi-agent Unmanned Surface 3 | Vehicles Containment Control Technology based on deep reinforcement learning and optimization theory -------------------------------------------------------------------------------- /documents/prove.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/3020663206/Optimal-Controller-for-Multiagent-USVs/47b8ebf629d97afb5d32b1f361384c8676f16358/documents/prove.pdf -------------------------------------------------------------------------------- /problem1/adp_drl_nn.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | 3 | import model_def 4 | 5 | eta_c_1 = 0.1 6 | eta_a_1 = 0.3 7 | eta_c_2 = 0.01 8 | eta_a_2 = 0.4 9 | 10 | zeta_1 = 10 11 | zeta_2 = 14 12 | 13 | class RBFN(object): 14 | 15 | def __init__(self, hidden_nums, output_nums): #还有一些超参数可能需要初始化 16 | self.hidden_nums = hidden_nums 17 | self.output_nums = output_nums 18 | self.feature_nums = 0 19 | self.sample_nums = 0 20 | self.gaussian_kernel_width = 0 # 高斯核宽度 21 | self.hiddencenters = 0 22 | self.hiddenoutputs = 0 23 | self.hiddenoutputs_expand = 0 24 | self.linearweights = 0 25 | self.finaloutputs = 0 26 | 27 | def init(self): 28 | gaussian_kernel_width = np.random.random((self.hiddencenters, 1)) #待修改 29 | hiddencenters = np.random.random((self.hidden_nums, self.feature_nums)) #待修改 30 | linearweights = np.random.random((self.hidden_nums + 1, self.output_nums)) #待修改 31 | return gaussian_kernel_width, hiddencenters, linearweights 32 | 33 | def forward(self, inputs): 34 | self.sample_nums, self.feature_nums = inputs.shape 35 | self.gaussian_kernel_width, self.hiddencenters, self.linearweights = self.init() 36 | self.hiddenoutputs = self.guass_change(self.gaussian_kernel_width, inputs, self.hiddencenters) 37 | self.hiddenoutputs_expand = self.add_intercept(self.hiddenoutputs) 38 | self.finaloutputs = np.dot(self.hiddenoutputs_expand, self.linearweights) 39 | 40 | def guass_function(self, gaussian_kernel_width, inputs, hiddencenters_i): 41 | return np.exp(-np.linalg.norm((inputs-hiddencenters_i), axis=1)**2/(2*gaussian_kernel_width**2)) 42 | 43 | def guass_change(self, gaussian_kernel_width, inputs, hiddencenters): 44 | hiddenresults = np.zeros((self.sample_nums, len(hiddencenters))) 45 | for i in range(len(hiddencenters)): 46 | hiddenresults[:,i] = self.guass_function(gaussian_kernel_width[i], inputs, hiddencenters[i]) 47 | return hiddenresults 48 | 49 | def add_intercept(self, hiddenoutputs): 50 | return np.hstack((hiddenoutputs, np.ones((self.sample_nums,1)))) 51 | 52 | class Critic1_NN(RBFN): 53 | 54 | def __init__(self, hidden_nums, output_nums): 55 | super().__init__(hidden_nums, output_nums) 56 | self.varpi_super = 0 57 | self.linearweights_last = 0 58 | 59 | def backward(self, certain_model, actor_1_nn): 60 | self.varpi_super = -self.hiddenoutputs_expand * [ 61 | certain_model.a * zeta_1 * certain_model.z_1 + 62 | 1 / 2 * certain_model.a * actor_1_nn.finaloutputs + certain_model.lambda_2] 63 | self.linearweights_last = self.linearweights 64 | reg_1 = (-2 * zeta_1 * certain_model.z_1.T * certain_model.lambda_2) 65 | reg_2 = -((certain_model.a * zeta_1 * zeta_1 - 1) * certain_model.z_1.T * certain_model.z_1) 66 | reg_3 = (1 / 4 * certain_model.a * actor_1_nn.finaloutputs * actor_1_nn.finaloutputs.T) 67 | reg_4 = (self.varpi_super.T * self.linearweights_last) 68 | self.linearweights += (-eta_c_1 / (1 + self.varpi_super * self.varpi_super.T) * self.varpi_super) * (reg_1 + reg_2 + reg_3 +reg_4) 69 | 70 | class Actor1_NN(RBFN): 71 | 72 | def __init__(self, hidden_nums, output_nums): 73 | super().__init__(hidden_nums, output_nums) 74 | self.varpi_super = 0 75 | self.linearweights_last = 0 76 | 77 | def backward(self, certain_model, critic_1_nn): 78 | self.varpi_super = -self.hiddenoutputs_expand * [ 79 | certain_model.a * zeta_1 * certain_model.z_1 + 80 | 1 / 2 * certain_model.a * self.finaloutputs + certain_model.lambda_2] 81 | self.linearweights_last = self.linearweights 82 | reg_1 = (1 / 2 * self.hiddenoutputs_expand.T * certain_model.z_1) 83 | reg_2 = ((eta_c_1 / (4 * (1 + self.varpi_super * self.varpi_super.T))) * self.hiddenoutputs_expand * self.hiddenoutputs_expand.T * self.linearweights_last * self.varpi_super.T * critic_1_nn.linearweights_last) 84 | reg_3 = (-eta_a_1 * self.hiddenoutputs_expand * self.hiddenoutputs_expand.T * self.linearweights_last) 85 | self.linearweights += (reg_1 + reg_2 + reg_3) 86 | 87 | class Critic2_NN(RBFN): 88 | 89 | def __init__(self, hidden_nums, output_nums): 90 | super().__init__(hidden_nums, output_nums) 91 | self.varpi_sub = 0 92 | self.linearweights_last = 0 93 | 94 | def backward(self, certain_model, actor_2_nn, a_hat_dot): 95 | self.varpi_sub = -self.hiddenoutputs_expand * [certain_model.function_V - zeta_2 * certain_model.z_2 96 | - 1 / 2 * actor_2_nn.finaloutputs - a_hat_dot] 97 | self.linearweights_last = self.linearweights 98 | reg_1 = 2 * zeta_2 * certain_model.z_2.T * (certain_model.function_V - a_hat_dot) 99 | reg_2 = - (zeta_2 * zeta_2 - 1) * certain_model.z_2.T * certain_model.z_2 100 | reg_3 = 1 / 4 * actor_2_nn.finaloutputs * actor_2_nn.finaloutputs.T 101 | reg_4 = self.varpi_sub.T * self.linearweights_last 102 | self.linearweights += (-eta_c_2 / (1 + self.varpi_sub * self.varpi_sub.T) * self.varpi_sub) * (reg_1 + reg_2 + reg_3 +reg_4) 103 | 104 | class Actor2_NN(RBFN): 105 | 106 | def __init__(self, hidden_nums, output_nums): 107 | super().__init__(hidden_nums, output_nums) 108 | self.varpi_sub = 0 109 | self.linearweights_last = 0 110 | 111 | def backward(self, certain_model, critic_2_nn, a_hat_dot): 112 | self.varpi_sub = -self.hiddenoutputs_expand * [certain_model.function_V - zeta_2 * certain_model.z_2 113 | - 1 / 2 * self.finaloutputs - a_hat_dot] 114 | self.linearweights_last = self.linearweights 115 | reg_1 = (1 / 2 * self.hiddenoutputs_expand.T * certain_model.z_2) 116 | reg_2 = ((eta_c_2 / (4 * (1 + self.varpi_sub * self.varpi_sub.T))) * self.hiddenoutputs_expand * self.hiddenoutputs_expand.T * self.linearweights_last * self.varpi_sub.T * critic_2_nn.linearweights_last) 117 | reg_3 = (-eta_a_2 * self.hiddenoutputs_expand * self.hiddenoutputs_expand.T * self.linearweights_last) 118 | self.linearweights += (reg_1 + reg_2 + reg_3) -------------------------------------------------------------------------------- /problem1/agent_def.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | from math import pi, atan2, sin, cos 3 | 4 | 5 | def get_abs_min_key(dct): 6 | return min(dct, key=lambda k: abs(dct[k])) 7 | 8 | def get_abs_max_key(dct): 9 | return max(dct, key=lambda k: abs(dct[k])) 10 | 11 | 12 | class Agent(object): 13 | def __init__(self, id, pos_x, pos_y): 14 | self.front_neighbor = None 15 | self.back_neighbor = None 16 | self.id = id 17 | self.pos_x = pos_x 18 | self.pos_y = pos_y 19 | #self.angle = angle 20 | self.front_neighbors_sets = {} 21 | self.back_neighbors_sets = {} 22 | self.front_neighbors_sets_1 = {} 23 | self.back_neighbors_sets_1 = {} 24 | self.front_neighbors_sets_2 = {} 25 | self.back_neighbors_sets_2 = {} 26 | 27 | def get_distance_and_angle(self, target_x, target_y): 28 | self.distance = np.sqrt((target_x - self.pos_x) * (target_x - self.pos_x) + (target_y - self.pos_y) * (target_y - self.pos_y)) 29 | self.angle = atan2(self.pos_y - target_y, self.pos_x - target_x) 30 | self.Q_transform = np.array(([cos(self.angle), -sin(self.angle)], [sin(self.angle), cos(self.angle)])) 31 | 32 | def get_sametype_neighbors(self, agents): 33 | self.front_neighbors_sets = {} 34 | self.back_neighbors_sets = {} 35 | self.front_neighbors_sets_1 = {} 36 | self.back_neighbors_sets_1 = {} 37 | self.front_neighbors_sets_2 = {} 38 | self.back_neighbors_sets_2 = {} 39 | for agent in agents: 40 | if agent.id != self.id: 41 | angle_diff = self.angle - agent.angle 42 | if self.angle >= 0: 43 | if self.angle - pi <= angle_diff <= 0: 44 | self.front_neighbors_sets_1[agent.id] = angle_diff 45 | elif pi < angle_diff <= pi + self.angle: 46 | self.front_neighbors_sets_2[agent.id] = angle_diff 47 | elif 0 < angle_diff < pi: 48 | self.back_neighbors_sets[agent.id] = angle_diff 49 | else: 50 | if -pi <= angle_diff <= 0: 51 | self.front_neighbors_sets[agent.id] = angle_diff 52 | elif self.angle - pi <= angle_diff < -pi: 53 | self.back_neighbors_sets_2[agent.id] = angle_diff 54 | elif 0 < angle_diff <= pi + self.angle: 55 | self.back_neighbors_sets_1[agent.id] = angle_diff 56 | if self.angle >= 0: 57 | if self.front_neighbors_sets_1 == {}: 58 | self.front_neighbor = get_abs_max_key(self.front_neighbors_sets_2) 59 | else: 60 | self.front_neighbor = get_abs_min_key(self.front_neighbors_sets_1) 61 | self.back_neighbor = get_abs_min_key(self.back_neighbors_sets) 62 | elif self.angle < 0: 63 | if self.back_neighbors_sets_1 == {}: 64 | self.back_neighbor = get_abs_max_key(self.back_neighbors_sets_2) 65 | else: 66 | self.back_neighbor = get_abs_min_key(self.back_neighbors_sets_1) 67 | self.front_neighbor = get_abs_min_key(self.front_neighbors_sets) 68 | return self.front_neighbor,self.back_neighbor 69 | 70 | if __name__ == "__main__": 71 | agents = [] 72 | agents.append(Agent(0, 1, 7)) 73 | agents.append(Agent(1, 5, 1)) 74 | agents.append(Agent(2, 2, 8)) 75 | 76 | for agent_i in agents: 77 | agent_i.get_distance_and_angle(0, 0) 78 | for agent_i in agents: 79 | agent_i.get_sametype_neighbors(agents) 80 | print(agent_i.get_sametype_neighbors(agents)) 81 | print(f"Agent {agent_i.id} front neighbors: {agent_i.front_neighbor}") 82 | print(f"Agent {agent_i.id} back neighbors: {agent_i.back_neighbor}") -------------------------------------------------------------------------------- /problem1/main.py: -------------------------------------------------------------------------------- 1 | # This is a sample Python script. 2 | 3 | # Press Shift+F10 to execute it or replace it with your code. 4 | # Press Double Shift to search everywhere for classes, files, tool windows, actions, and settings. 5 | 6 | 7 | def print_hi(name): 8 | # Use a breakpoint in the code line below to debug your script. 9 | print(f'Hi, {name}') # Press Ctrl+F8 to toggle the breakpoint. 10 | 11 | 12 | # Press the green button in the gutter to run the script. 13 | if __name__ == '__main__': 14 | print_hi('PyCharm') 15 | 16 | # See PyCharm help at https://www.jetbrains.com/help/pycharm/ 17 | -------------------------------------------------------------------------------- /problem1/model_def.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | from math import pi, atan2, sin, cos 3 | 4 | import agent_def 5 | import adp_drl_nn 6 | 7 | M_11 = 1.956 8 | M_22 = 2.405 9 | M_33 = 0.043 10 | D_11 = 2.436 11 | D_22 = 12.992 12 | D_33 = 0.0564 13 | 14 | rho_0 = 3 15 | R_rewardweights = np.eye(3) 16 | 17 | T_changeNN = 0.1 18 | T_changeState = 0.01 19 | 20 | class Hunter(agent_def.Agent): 21 | 22 | def __init__(self, id, pos_x, pos_y, orientation, speed_u, speed_v, speed_r): 23 | super().__init__(id, pos_x, pos_y) 24 | self.orientation = orientation 25 | self.speed_u = speed_u 26 | self.speed_v = speed_v 27 | self.speed_r = speed_r 28 | self.vector_V = np.array([self.speed_u, self.speed_v, self.speed_r]).T 29 | self.a = 0 30 | self.u_hat = [0, 0] 31 | self.optimal_V_hat = 0 32 | self.optimal_V_hat_last = 0 33 | self.optimal_V_hat_dot = 0 34 | 35 | def change_position(self, tau_1, tau_2): 36 | u_next = (tau_1 - D_11 * self.speed_u + M_22 * self.speed_v * self.speed_r) / M_11 * T_changeState 37 | v_next = (-D_22 * self.speed_v - M_11 * self.speed_u * self.speed_r) / M_22 * T_changeState 38 | r_next = (tau_2 - D_33 * self.speed_r - (M_22 - M_11) * self.speed_u * self.speed_v) / M_33 * T_changeState 39 | # 运动学模型 （1） 40 | self.speed_u += u_next 41 | self.speed_v += v_next 42 | self.speed_r += r_next 43 | self.pos_x += (self.speed_u * cos(self.orientation) - self.speed_v * sin(self.orientation)) * T_changeState 44 | self.pos_y += (self.speed_u * sin(self.orientation) + self.speed_v * cos(self.orientation)) * T_changeState 45 | self.orientation += self.speed_r * T_changeState 46 | self.function_V = np.array([(-D_11 * self.speed_u + M_22 * self.speed_v * self.speed_r) / M_11], 47 | [(-D_22 * self.speed_v - M_11 * self.speed_u * self.speed_r) / M_22], 48 | [(-D_33 * self.speed_r - (M_22 - M_11) * self.speed_u * self.speed_v) / M_33]) 49 | 50 | def calculate_super_error(self, angle_front, angle_behind, rho_front, rho_behind, u_front, u_behind, v_front, v_behind, v_target_x, v_target_y): 51 | 52 | self.z_1 = (self.distance - rho_0) ** 2 + (2 * self.angle - angle_front - angle_behind) ** 2 + (self.orientation + pi / 2 - self.angle) 53 | 54 | self.alpha_1 = 2 * (self.distance - rho_0) * (np.array([[cos(self.angle), sin(self.angle)]])) 55 | #print(self.alpha_1) 56 | 57 | self.beta_1 = (4 * (2 * self.angle - angle_front - angle_behind) * (np.array([[-sin(self.angle), cos(self.angle)]]))) / self.distance 58 | self.beta_2 = (2 * (2 * self.angle - angle_front - angle_behind) * (np.array([[-sin(angle_front), cos(angle_behind)]]))) / rho_front 59 | self.beta_3 = (2 * (2 * self.angle - angle_front - angle_behind) * (np.array([[-sin(angle_behind), cos(angle_behind)]]))) / rho_behind 60 | #print(self.beta_1) 61 | #print(self.beta_2) 62 | #print(self.beta_3) 63 | self.gamma_1 = 2 * (self.orientation + pi / 2 - self.angle) 64 | self.gamma_2 = (2 * (self.orientation + pi / 2 - self.angle) * (np.array([[-sin(self.angle), cos(self.angle)]]))) / self.distance 65 | #print(self.gamma_1) 66 | #print(self.gamma_2) 67 | Q_transform_front = np.array([[cos(angle_front), -sin(angle_front)], [sin(angle_front), cos(angle_front)]]) 68 | Q_transform_behind = np.array([[cos(angle_behind), -sin(angle_behind)], [sin(angle_behind), cos(angle_behind)]]) 69 | 70 | self.delta = self.beta_2 * Q_transform_front * np.array([u_front, v_front]) + self.beta_3 * Q_transform_behind * np.array([u_behind, v_behind]) 71 | 72 | self.lambda_1 = np.array([(self.alpha_1 + self.beta_1 - self.gamma_1) * self.Q_transform, self.gamma_1]) 73 | # print(self.Q_transform) 74 | # print(self.Q_transform.shape) 75 | # print((self.alpha_1 + self.beta_1 - self.gamma_1)) 76 | # print((self.alpha_1 + self.beta_1 - self.gamma_1).shape) 77 | # print(((self.alpha_1 + self.beta_1 - self.gamma_1) * self.Q_transform)) 78 | # print(((self.alpha_1 + self.beta_1 - self.gamma_1) * self.Q_transform).shape) 79 | # print(self.gamma_1) 80 | # print(self.lambda_1) 81 | # print(self.lambda_1.shape) 82 | self.lambda_2 = (self.alpha_1 + self.beta_1 - self.beta_2 - self.beta_3) * np.array([v_target_x, v_target_y]) 83 | 84 | self.dot_z_1 = self.lambda_1 * np.array([self.speed_u, self.speed_v, self.speed_r]) * self.lambda_2 85 | 86 | self.a = self.lambda_1 * np.linalg.inv(R_rewardweights) * self.lambda_1.T 87 | 88 | return self.z_1, self.dot_z_1 89 | 90 | def calculate_virtual_opitmal(self, actor_1_nn_outputs): 91 | self.optimal_V_hat_last = self.optimal_V_hat 92 | self.optimal_V_hat = np.linalg.inv(R_rewardweights) * self.lambda_1.T * ( 93 | -adp_drl_nn.zeta_1 * self.z_1 - 1 / 2 * actor_1_nn_outputs) 94 | self.optimal_V_hat_dot = (self.optimal_V_hat - self.optimal_V_hat_last) / T_changeNN 95 | 96 | def calculate_sub_error(self): 97 | 98 | self.z_2 = self.vector_V - self.optimal_V_hat 99 | return self.z_2 100 | 101 | def calculate_actual_opitmal(self, actor_2_nn_outputs): 102 | self.u_hat = -adp_drl_nn.zeta_2 * self.z_2 - 1 / 2 * actor_2_nn_outputs 103 | 104 | 105 | class Invader(agent_def.Agent): 106 | 107 | pos_x, pos_y, orientation = 0, 0, 0 108 | speed_x_axis, speed_y_axis = 0, 0 109 | speed_u, speed_v = 0, 0 110 | 111 | def __init__(self, id, pos_x, pos_y, orientation): 112 | super().__init__(id, pos_x, pos_y) 113 | self.orientation = orientation 114 | 115 | def change_speed(self, speed_x_axis, speed_y_axis): 116 | self.speed_x_axis = speed_x_axis 117 | self.speed_y_axis = speed_y_axis 118 | self.ang = atan2(speed_y_axis, speed_x_axis) 119 | self.speed_u = speed_x_axis * cos(self.orientation) + speed_y_axis * sin(self.orientation) 120 | self.speed_v = -speed_x_axis * sin(self.orientation) + speed_y_axis * cos(self.orientation) 121 | 122 | def change_position(self): 123 | self.pos_x += self.speed_x_axis * T_changeState 124 | self.pos_y += self.speed_y_axis * T_changeState -------------------------------------------------------------------------------- /problem1/rbf.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.nn as nn 3 | import numpy as np 4 | 5 | class RBFLayer(nn.Module): 6 | def __init__(self, input_dim, num_centers, sigma=1.0): 7 | super(RBFLayer, self).__init__() 8 | 9 | self.num_centers = num_centers 10 | self.sigma = sigma 11 | self.centers = nn.Parameter(torch.randn(num_centers, input_dim)) 12 | self.linear = nn.Linear(num_centers, 3) 13 | 14 | def forward(self, x): 15 | # 计算径向基函数的输出 16 | x = x.unsqueeze(1).expand(x.size(0), self.num_centers, x.size(-1)) 17 | c = self.centers.unsqueeze(0).expand(x.size(0), self.num_centers, x.size(-1)) 18 | distances = sum 19 | output = (-distances / (2 * self.sigma ** 2)).exp() 20 | 21 | # 使用线性层计算最终输出 22 | output = self.linear(output) 23 | 24 | return output 25 | 26 | # 创建数据集 27 | x_train = np.random.rand(100, 1) 28 | y_train = np.sin(x_train * np.pi) + np.random.normal(0, 0.1, (100, 1)) 29 | 30 | # 将数据转换为 PyTorch 张量 31 | x_train = torch.tensor(x_train, dtype=torch.float) 32 | y_train = torch.tensor(y_train, dtype=torch.float) 33 | 34 | # 创建模型 35 | model = RBFLayer(1, 72) 36 | 37 | # 定义损失函数 38 | criterion = nn.MSELoss() 39 | 40 | # 定义优化器 41 | class CustomOptimizer(torch.optim.Optimizer): 42 | def __init__(self, params, lr=1e-3): 43 | defaults = dict(lr=lr) 44 | super(CustomOptimizer, self).__init__(params, defaults) 45 | 46 | def step(self, closure=None): 47 | loss = None 48 | if closure is not None: 49 | loss = closure() 50 | 51 | for group in self.param_groups: 52 | for p in group['params']: 53 | if p.grad is None: 54 | continue 55 | grad = p.grad.data 56 | state = self.state[p] 57 | lr = group['lr'] 58 | if len(state) == 0: 59 | state['step'] = 0 60 | state['sum'] = torch.zeros_like(p.data) 61 | state['step'] += 1 62 | state['sum'] += grad ** 2 63 | rms = state['sum'] / state['step'] 64 | p.data -= lr * grad / torch.sqrt(rms + 1e-8) 65 | 66 | return loss 67 | 68 | optimizer = CustomOptimizer(model.parameters(), lr=0.1) 69 | 70 | # 训练模型 71 | for epoch in range(1000): 72 | # 前向传播 73 | y_pred = model(x_train) 74 | 75 | # 计算损失函数 76 | loss = criterion(y_pred, y_train) 77 | 78 | # 反向传播和权重更新 79 | optimizer.zero_grad() 80 | loss.backward() 81 | optimizer.step() 82 | 83 | # 输出每 100 次迭代后的损失值 84 | if epoch % 100 == 0: 85 | print(f"Epoch {epoch}, Loss: {loss.item():.4f}") 86 | #print(model.linear.weight) 87 | 88 | # 测试模型 89 | x_test = torch.linspace(0, 1, 100).unsqueeze(1) 90 | y_test = model(x_test) 91 | 92 | # 绘制预测结果和原始数据的比较图 93 | import matplotlib.pyplot as plt 94 | plt.plot(x_train.numpy(), y_train.numpy(), 'o', label='Original data') 95 | plt.plot(x_test.numpy(), y_test.detach().numpy(), label='Fitted line') 96 | plt.legend() 97 | plt.show() 98 | 99 | 100 | -------------------------------------------------------------------------------- /problem1/rbf_network.py: -------------------------------------------------------------------------------- 1 | import torch, random 2 | import torch.nn as nn 3 | import torch.optim as optim 4 | 5 | torch.manual_seed(42) 6 | 7 | 8 | class RBFN_type1(nn.Module): 9 | """ 10 | 以高斯核作为径向基函数 11 | """ 12 | 13 | def __init__(self, centers, n_out=3): 14 | """ 15 | :param centers: shape=[center_num,data_dim] 16 | :param n_out: 17 | """ 18 | super(RBFN_type1, self).__init__() 19 | self.n_out = n_out 20 | self.num_centers = centers.size(0) # 隐层节点的个数 21 | self.dim_centure = centers.size(1) # 22 | self.centers = nn.Parameter(centers) 23 | # self.beta = nn.Parameter(torch.ones(1, self.num_centers), requires_grad=True) 24 | self.beta = torch.ones(1, self.num_centers) * 10 25 | # 对线性层的输入节点数目进行了修改 26 | self.linear = nn.Linear(self.num_centers + self.dim_centure, self.n_out, bias=True) 27 | self.initialize_weights() # 创建对象时自动执行 28 | 29 | def kernel_fun(self, batches): 30 | n_input = batches.size(0) # number of inputs 31 | A = self.centers.view(self.num_centers, -1).repeat(n_input, 1, 1) 32 | B = batches.view(n_input, -1).unsqueeze(1).repeat(1, self.num_centers, 1) 33 | C = torch.exp(-self.beta.mul((A - B).pow(2).sum(2, keepdim=False))) 34 | return C 35 | 36 | def forward(self, batches): 37 | radial_val = self.kernel_fun(batches) 38 | class_score = self.linear(torch.cat([batches, radial_val], dim=1)) 39 | return class_score 40 | 41 | def initialize_weights(self, ): 42 | """ 43 | 网络权重初始化 44 | :return: 45 | """ 46 | for m in self.modules(): 47 | if isinstance(m, nn.Conv2d): 48 | m.weight.data.normal_(0, 0.02) 49 | m.bias.data.zero_() 50 | elif isinstance(m, nn.ConvTranspose2d): 51 | m.weight.data.normal_(0, 0.02) 52 | m.bias.data.zero_() 53 | elif isinstance(m, nn.Linear): 54 | m.weight.data.normal_(0, 0.02) 55 | m.bias.data.zero_() 56 | 57 | def print_network(self): 58 | num_params = 0 59 | for param in self.parameters(): 60 | num_params += param.numel() 61 | print(self) 62 | print('Total number of parameters: %d' % num_params) 63 | 64 | class RBFN_type2(nn.Module): 65 | """ 66 | 以高斯核作为径向基函数 67 | """ 68 | 69 | def __init__(self, centers, n_out=3): 70 | """ 71 | :param centers: shape=[center_num,data_dim] 72 | :param n_out: 73 | """ 74 | super(RBFN_type2, self).__init__() 75 | self.n_out = n_out 76 | self.num_centers = centers.size(0) # 隐层节点的个数 77 | self.dim_centure = centers.size(1) # 78 | self.centers = nn.Parameter(centers) 79 | # self.beta = nn.Parameter(torch.ones(1, self.num_centers), requires_grad=True) 80 | self.beta = torch.ones(1, self.num_centers) * 10 81 | # 对线性层的输入节点数目进行了修改 82 | self.linear = nn.Linear(self.num_centers + self.dim_centure, self.n_out, bias=True) 83 | self.initialize_weights() # 创建对象时自动执行 84 | 85 | def kernel_fun(self, batches): 86 | n_input = batches.size(0) # number of inputs 87 | A = self.centers.view(self.num_centers, -1).repeat(n_input, 1, 1) 88 | B = batches.view(n_input, -1).unsqueeze(1).repeat(1, self.num_centers, 1) 89 | C = torch.exp(-self.beta.mul((A - B).pow(2).sum(2, keepdim=False))) 90 | return C 91 | def forward(self, batches): 92 | radial_val = self.kernel_fun(batches) 93 | class_score = self.linear(torch.cat([batches, radial_val], dim=1)) 94 | return class_score 95 | 96 | def initialize_weights(self, ): 97 | """ 98 | 网络权重初始化 99 | :return: 100 | """ 101 | for m in self.modules(): 102 | if isinstance(m, nn.Conv2d): 103 | m.weight.data.normal_(0, 0.02) 104 | m.bias.data.zero_() 105 | elif isinstance(m, nn.ConvTranspose2d): 106 | m.weight.data.normal_(0, 0.02) 107 | m.bias.data.zero_() 108 | elif isinstance(m, nn.Linear): 109 | m.weight.data.normal_(0, 0.02) 110 | m.bias.data.zero_() 111 | 112 | def print_network(self): 113 | num_params = 0 114 | for param in self.parameters(): 115 | num_params += param.numel() 116 | print(self) 117 | print('Total number of parameters: %d' % num_params) 118 | 119 | class RBFN_type3(torch.nn.Module): 120 | def __init__(self, centers, n_out=3): 121 | super(RBFN_type3, self).__init__() 122 | self.n_out = n_out 123 | self.num_centers = centers.size(0) 124 | self.centers = torch.nn.Parameter(centers) 125 | self.beta = torch.nn.Parameter(torch.ones(1, self.num_centers)) 126 | self.linear = torch.nn.Linear(self.num_centers + n_out, self.n_out) 127 | self.initialize_weights() 128 | 129 | def kernel_fun(self, batches): 130 | n_input = batches.size(0) 131 | c = self.centers.view(self.num_centers, -1).repeat(n_input, 1, 1) # torch.Size([500, 500, 1]) 132 | x = batches.view(n_input, -1).unsqueeze(1).repeat(1, self.num_centers, 1) # torch.Size([500, 500, 1]) 133 | radial_val = torch.exp(-self.beta.mul((c - x).pow(2).sum(2))) 134 | return radial_val 135 | 136 | def forward(self, x): 137 | radial_val = self.kernel_fun(x) 138 | out = self.linear(torch.cat([x, radial_val], dim=1)) 139 | return out, torch.cat([x, radial_val]) 140 | 141 | def initialize_weights(self): 142 | for m in self.modules(): 143 | if isinstance(m, torch.nn.Conv2d): 144 | m.weight.data.normal_(0, 0.2) 145 | m.bias.data.zero_() 146 | elif isinstance(m, torch.nn.ConvTranspose2d): 147 | m.weight.data.normal_(0, 0.2) 148 | m.bias.data.zero_() 149 | elif isinstance(m, torch.nn.Linear): 150 | m.weight.data.normal_(0, 0.2) 151 | m.bias.data.zero_() 152 | 153 | 154 | # centers = torch.rand((5,8)) 155 | # rbf_net = RBFN(centers) 156 | # rbf_net.print_network() 157 | # rbf_net.initialize_weights() 158 | 159 | 160 | if __name__ == "__main__": 161 | data = torch.tensor([[0.25, 0.75], [0.75, 0.75], [0.25, 0.5], [0.5, 0.5], [0.75, 0.5], 162 | [0.25, 0.25], [0.75, 0.25], [0.5, 0.125], [0.75, 0.125]], dtype=torch.float32) 163 | label = torch.tensor([[-1, 1, -1], [1, -1, -1], [-1, -1, 1], [-1, -1, 1], [-1, -1, 1], 164 | [1, -1, -1], [-1, 1, -1], [-1, 1, -1], [1, -1, -1]], dtype=torch.float32) 165 | print(data.size()) 166 | 167 | centers = data[0:8, :] 168 | rbf = RBFN_type1(centers, 3) 169 | params = rbf.parameters() 170 | loss_fn = torch.nn.MSELoss() 171 | optimizer = torch.optim.SGD(params, lr=0.1, momentum=0.9) 172 | 173 | for i in range(10000): 174 | optimizer.zero_grad() 175 | 176 | y = rbf.forward(data) 177 | loss = loss_fn(y, label) 178 | loss.backward() 179 | optimizer.step() 180 | print(i, "\t", loss.data) 181 | 182 | # 加载使用 183 | y = rbf.forward(data) 184 | print(y.data) 185 | print(label.data) -------------------------------------------------------------------------------- /problem1/reg.py: -------------------------------------------------------------------------------- 1 | hunter_0_critic_1 = adp_drl_nn.Critic1_NN(center_numbers) 2 | hunter_0_actor_1 = adp_drl_nn.Actor1_NN(center_numbers) 3 | hunter_0_critic_2 = adp_drl_nn.Critic2_NN(center_numbers) 4 | hunter_0_actor_2 = adp_drl_nn.Actor2_NN(center_numbers) 5 | 6 | hunter_1_critic_1 = adp_drl_nn.Critic1_NN(center_numbers) 7 | hunter_1_actor_1 = adp_drl_nn.Actor1_NN(center_numbers) 8 | hunter_1_critic_2 = adp_drl_nn.Critic2_NN(center_numbers) 9 | hunter_1_actor_2 = adp_drl_nn.Actor2_NN(center_numbers) 10 | 11 | hunter_2_critic_1 = adp_drl_nn.Critic1_NN(center_numbers) 12 | hunter_2_actor_1 = adp_drl_nn.Actor1_NN(center_numbers) 13 | hunter_2_critic_2 = adp_drl_nn.Critic2_NN(center_numbers) 14 | hunter_2_actor_2 = adp_drl_nn.Actor2_NN(center_numbers) 15 | 16 | all_critic_1.append(hunter_0_critic_1) 17 | all_critic_1.append(hunter_1_critic_1) 18 | all_critic_1.append(hunter_2_critic_1) 19 | 20 | all_actor_1.append(hunter_0_actor_1) 21 | all_actor_1.append(hunter_1_actor_1) 22 | all_actor_1.append(hunter_2_actor_1) 23 | 24 | all_critic_2.append(hunter_0_critic_2) 25 | all_critic_2.append(hunter_1_critic_2) 26 | all_critic_2.append(hunter_2_critic_2) 27 | 28 | all_actor_2.append(hunter_0_actor_2) 29 | all_actor_2.append(hunter_1_actor_2) 30 | all_actor_2.append(hunter_2_actor_2) -------------------------------------------------------------------------------- /problem1/test_unit1.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import torch.nn as nn 3 | import torch.optim as optim 4 | import matplotlib.pyplot as plt 5 | import time 6 | """ 7 | 参考：https://goodgoodstudy.blog.csdn.net/article/details/105756137 8 | """ 9 | 10 | def SaveImage(label,pre,path): 11 | label = label.view(-1).cpu().detach().numpy() 12 | pre = pre.view(-1).cpu().detach().numpy() 13 | plt.rcParams['font.sans-serif'] = 'KaiTi' 14 | plt.rcParams['axes.unicode_minus'] = False 15 | fig = plt.figure(dpi=400) 16 | ax = fig.add_subplot(111) 17 | ax.plot(label, color='blue', label="实际值") 18 | ax.plot(pre, color='red', linestyle='--', label='拟合值') 19 | ax.legend() 20 | fig.savefig(path, dpi=400) 21 | 22 | class RBF(nn.Module): 23 | def __init__(self,centers,n_out=1): 24 | super(RBF, self).__init__() 25 | self.n_out = n_out 26 | self.num_centers = centers.size(0) 27 | 28 | self.centers = nn.Parameter(centers) 29 | self.beta = nn.Parameter(torch.ones(1,self.num_centers)) 30 | self.linear = nn.Linear(self.num_centers+n_out,self.n_out) 31 | self.initialize_weights() 32 | 33 | def kernel_fun(self,batches): 34 | n_input = batches.size(0) 35 | c = self.centers.view(self.num_centers,-1).repeat(n_input,1,1)# torch.Size([500, 500, 1]) 36 | x = batches.view(n_input,-1).unsqueeze(1).repeat(1,self.num_centers,1)# torch.Size([500, 500, 1]) 37 | radial_val = torch.exp(-self.beta.mul((c-x).pow(2).sum(2))) 38 | return radial_val 39 | 40 | def forward(self,x): 41 | # 计算径向基距离函数 42 | radial_val = self.kernel_fun(x) 43 | out = self.linear(torch.cat([x,radial_val],dim=1)) 44 | return out 45 | 46 | def initialize_weights(self): 47 | for m in self.modules(): 48 | if isinstance(m, nn.Conv2d): 49 | m.weight.data.normal_(0, 0.2) 50 | m.bias.data.zero_() 51 | elif isinstance(m, nn.ConvTranspose2d): 52 | m.weight.data.normal_(0, 0.2) 53 | m.bias.data.zero_() 54 | elif isinstance(m, nn.Linear): 55 | m.weight.data.normal_(0, 0.2) 56 | m.bias.data.zero_() 57 | 58 | 59 | num_centers = 72 60 | n_out = 3 61 | centers = torch.randn(num_centers,n_out) 62 | 63 | model = RBF(centers,n_out=3) 64 | optimizer = optim.Adam(model.parameters(),lr=0.1) 65 | loss_fun = nn.MSELoss() 66 | 67 | X_ = torch.linspace(-5,5,500).view(500,1) 68 | Y_ = torch.mul(1.1*(1-X_+X_.pow(2).mul(2)),torch.exp(X_.pow(2).mul(-0.5))) 69 | 70 | start = time.time() 71 | epochs = 1000 72 | for epoch in range(epochs): 73 | avg_loss = 0 74 | Y_pre = model(X_) 75 | loss = loss_fun(Y_pre,Y_) 76 | 77 | optimizer.zero_grad() 78 | loss.backward() 79 | optimizer.step() 80 | 81 | print("epoch:{}\t loss:{:>.9}".format(epoch+1,loss.item())) 82 | end = time.time() 83 | print("time:",end-start) 84 | 85 | Y_pre = model(X_) 86 | SaveImage(Y_,Y_pre,"RBF.png") 87 | 88 | -------------------------------------------------------------------------------- /problem1/test_unit2.py: -------------------------------------------------------------------------------- 1 | import model_def 2 | import torch 3 | def num(): 4 | a = 1 5 | b = 2 6 | c = 2 7 | return a,b,c 8 | 9 | r = num() 10 | print(r[1]) 11 | print(torch.__version__) 12 | print(dir(torch.distributions)) -------------------------------------------------------------------------------- /problem1/test_unit3.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import matplotlib.pyplot as plt 3 | import os 4 | 5 | class RBFnetwork(object): 6 | def __init__(self, hidden_nums, r_w, r_c, r_sigma): 7 | self.h = hidden_nums #隐含层神经元个数 8 | self.w = 0 #线性权值 9 | self.c = 0 #神经元中心点 10 | self.sigma = 0 #高斯核宽度 11 | self.r = {"w":r_w, 12 | "c":r_c, 13 | "sigma":r_sigma} #参数迭代的学习率 14 | self.errList = [] #误差列表 15 | self.n_iters = 0 #实际迭代次数 16 | self.tol = 1.0e-5 #最大容忍误差 17 | self.X = 0 #训练集特征 18 | self.y = 0 #训练集结果 19 | self.n_samples = 0 #训练集样本数量 20 | self.n_features = 0 #训练集特征数量 21 | 22 | #训练 23 | def train(self, X, y, iters): 24 | self.X = X 25 | self.y = y.reshape(-1,1) 26 | self.n_samples, self.n_features = X.shape 27 | sigma, c, w = self.init() #初始化参数 28 | for i in range(iters): 29 | ##正向计算过程 30 | hi_output = self.change(sigma,X,c) #隐含层输出(m,h)，即通过径向基函数的转换 31 | yi_input = self.addIntercept(hi_output) #输出层输入(m,h+1)，因为是线性加权，故将偏置加入 32 | yi_output = np.dot(yi_input, w) #输出预测值(m,1) 33 | error = self.calSSE(yi_output, y) #计算误差 34 | if error < self.tol: 35 | break 36 | self.errList.append(error) #保存误差 37 | ##误差反向传播过程 38 | deltaw = np.dot(yi_input.T, (yi_output-y)) #(h+1,m)x(m,1) 39 | w -= self.r['w']*deltaw/self.n_samples 40 | deltasigma = np.divide(np.multiply(np.dot(np.multiply(hi_output,self.l2(X,c)).T, \ 41 | (yi_output-y)), w[:-1]), sigma**3) #(h,m)x(m,1) 42 | sigma -= self.r['sigma']*deltasigma/self.n_samples 43 | deltac1 = np.divide(w[:-1],sigma**2) #(h,1) 44 | deltac2 = np.zeros((1,self.n_features)) #(1,n) 45 | for j in range(self.n_samples): 46 | deltac2 += (yi_output-y)[j]*np.dot(hi_output[j], X[j]-c) 47 | deltac = np.dot(deltac1,deltac2) #(h,1)x(1,n) 48 | c -= self.r['c']*deltac/self.n_samples 49 | self.c = c 50 | self.w = w 51 | self.sigma = sigma 52 | self.n_iters = i 53 | 54 | #计算径向基距离函数 55 | def guass(self, sigma, X, ci): 56 | return np.exp(-np.linalg.norm((X-ci), axis=1)**2/(2*sigma**2)) 57 | 58 | #将原数据高斯转化成新数据 59 | def change(self, sigma, X, c): 60 | newX = np.zeros((self.n_samples, len(c))) 61 | for i in range(len(c)): 62 | newX[:,i] = self.guass(sigma[i], X, c[i]) 63 | return newX 64 | 65 | #初始化参数 66 | def init(self): 67 | sigma = np.random.random((self.h, 1)) #(h,1) 68 | c = np.random.random((self.h, self.n_features)) #(h,n) 69 | w = np.random.random((self.h+1, 1)) #(h+1,1) 70 | return sigma, c, w 71 | 72 | #给输出层的输入加一列截距项 73 | def addIntercept(self, X): 74 | return np.hstack((X,np.ones((self.n_samples,1)))) 75 | 76 | #计算整体误差 77 | def calSSE(self, prey, y): 78 | return 0.5*(np.linalg.norm(prey - y))**2 79 | 80 | #求L2范数的平方 81 | def l2(self, X, c): 82 | m,n = np.shape(X) 83 | newX = np.zeros((m, len(c))) 84 | for i in range(len(c)): 85 | newX[:,i] = np.linalg.norm((X-c[i]), axis=1)**2 86 | return newX 87 | 88 | #预测 89 | def predict(self, X): 90 | hi_output = self.change(self.sigma,X,self.c) #隐含层输出(m,h)，即通过径向基函数的转换 91 | yi_input = self.addIntercept(hi_output) #输出层输入(m,h+1)，因为是线性加权，故将偏置加入 92 | yi_output = np.dot(yi_input, self.w) #输出预测值(m,1) 93 | return yi_output 94 | 95 | if __name__ == "__main__": 96 | #拟合Hermit多项式 97 | X = np.linspace(-5, 5 , 500)[:, np.newaxis] 98 | y = np.multiply(1.1*(1-X+2*X**2),np.exp(-0.5*X**2)) 99 | rbf = RBFnetwork(50, 0.1, 0.2, 0.1) 100 | rbf.train(X, y, 1000) -------------------------------------------------------------------------------- /problem1/test_unit4.py: -------------------------------------------------------------------------------- 1 | # 使用深度强化学习实现无人艇集群合围算法 2 | 3 | import matplotlib.pyplot as plt 4 | from matplotlib.patches import Circle 5 | import numpy as np 6 | from math import pi, atan2, sqrt, sin, cos, atan 7 | import torch 8 | import model_def 9 | import rbf_network 10 | 11 | zeta_1 = 10 12 | zeta_2 = 14 13 | 14 | R = np.eye(3) 15 | 16 | eta_c_1 = 0.1 17 | eta_c_2 = 0.01 18 | eta_a_1 = 0.3 19 | eta_a_2 = 0.4 20 | num_centers = 72 21 | centers_c_1 = 0 22 | centers_a_1 = 0 23 | centers_c_2 = 0 24 | centers_a_2 = 0 25 | n_out_c_1 = 1 26 | n_out_c_2 = 3 27 | n_out_a_1 = 3 28 | n_out_a_2 = 2 29 | 30 | NumberofHunters = 3 31 | NumberofInvaders = 1 32 | 33 | #model_1 = model(0, 0, 0, 0, 0, 0) 34 | #model_2 = model(0, 0, 0, 0, 0, 0) 35 | #model_3 = model(0, 0, 0, 0, 0, 0) 36 | 37 | allhunters = [model_def.Hunter(0, 0, 0, 0, 0, 0) for i in range(model_def.NumberofHunters)] 38 | 39 | model_c_1 = rbf_network.RBFN_type1(model_def.centers_c_1, model_def.n_out_c_1) 40 | model_a_1 = rbf_network.RBFN_type1(model_def.centers_a_1, model_def.n_out_a_1) 41 | model_c_2 = rbf_network.RBFN_type1(model_def.centers_c_2, model_def.n_out_c_2) 42 | model_a_2 = rbf_network.RBFN_type1(model_def.centers_a_2, model_def.n_out_a_2) 43 | 44 | epochs = 500 45 | a_hat_dot = 0 46 | a_hat = 0 47 | 48 | for epoch in range(epochs): 49 | 50 | for i in range(model_def.NumberofHunters): 51 | # Step 1 52 | out_put_c_1 = model_c_1.forward(allhunters[i].z_1)[0] 53 | dot_v_1 = 2 * model_def.zeta_1 * allhunters[i].z_1 + out_put_c_1 54 | out_put_a_1 = model_a_1.forward(allhunters[i].z_1)[0] 55 | a_hat_last = a_hat 56 | a_hat = np.linalg.inv(model_def.R) * allhunters[i].lambda_1 * (-model_def.zeta_1 * allhunters[i].z_1 - 1 / 2 * out_put_a_1) 57 | omaga_c_1 = -model_c_1.forward(allhunters[i].z_1)[1] * [allhunters[i].a * model_def.zeta_1 * allhunters[i].z_1 + 1 / 2 * allhunters[i].a * out_put_a_1 + allhunters[i].lambda_2] 58 | # critic 网络更新 59 | model_c_1.linear.weight += (-model_def.gamma_c_1 / (1 + omaga_c_1 * omaga_c_1.T) * omaga_c_1 * [-2 * model_def.zeta_1 * allhunters[i].z_1.T * allhunters[i].lambda_2 60 | - (allhunters[i].a * model_def.zeta_1 * model_def.zeta_1 - 1) * allhunters[i].z_1.T * allhunters[i].z_1 + 1 / 4 * allhunters[i].a * out_put_a_1 * out_put_a_1.T + omaga_c_1.T * model_c_1.linear.weight]) 61 | # actor 网络更新 62 | model_a_1.linear.weight += (1 / 2 * model_c_1.forward(allhunters[i].z_1)[1].T * allhunters[i].z_1 + model_def.gamma_c_1 / (4 * (1 + omaga_c_1 * omaga_c_1.T)) 63 | * model_c_1.forward(allhunters[i].z_1)[1].T * model_c_1.forward(allhunters[i].z_1)[1] * model_a_1.linear.weight * omaga_c_1.T * model_c_1.linear.weight 64 | - model_def.gamma_a_1 * model_c_1.forward(allhunters[i].z_1)[1].T * model_c_1.forward(allhunters[i].z_1)[1] * model_a_1.linear.weight) 65 | 66 | # Step 2 67 | a_hat_dot = (a_hat - a_hat_last)/model_def.T_changeNN 68 | output_c_2 = model_c_1.forward(allhunters[i].z_2)[0] 69 | dot_v_2 = 2 * model_def.zeta_2 * allhunters[i].z_2 + output_c_2 70 | out_put_a_2 = model_a_2.forward(allhunters[i].z_2)[0] 71 | u_hat = -model_def.zeta_2 * allhunters[i].z_2 - 1 / 2 * out_put_a_2 72 | omaga_c_2 = -model_c_2.forward(allhunters[i].z_2)[1] * [allhunters[i].f_v - model_def.zeta_2 * allhunters[i].z_2 - 1 / 2 * out_put_a_2 - a_hat_dot] 73 | # critic 网络更新 74 | model_c_2.linear.weight += (-model_def.gamma_c_2 / (1 + omaga_c_2 * omaga_c_2.T) * omaga_c_2 * [2 * model_def.zeta_2 * allhunters[i].z_2.T * (allhunters[i].f_v - a_hat_dot) 75 | - (model_def.zeta_2 * model_def.zeta_2 - 1) * allhunters[i].z_2.T * allhunters[i].z_2 + 1 / 4 * out_put_a_2 * out_put_a_2.T + omaga_c_2.T * model_c_2.linear.weight]) 76 | # actor 网络更新 77 | model_a_2.linear.weight += (1 / 2 * model_c_2.forward(allhunters[i].z_2)[1].T * allhunters[i].z_2 + model_def.gamma_c_2 / (4 * (1 + omaga_c_2 * omaga_c_2.T)) 78 | * model_c_2.forward(allhunters[i].z_2)[1].T * model_c_2.forward(allhunters[i].z_2)[1] * model_a_2.linear.weight * omaga_c_2.T * model_c_2.linear.weight 79 | - model_def.gamma_a_2 * model_c_2.forward(allhunters[i].z_2)[1].T * model_c_2.forward(allhunters[i].z_2)[1] * model_a_2.linear.weight) 80 | 81 | 82 | 83 | 84 | 85 | -------------------------------------------------------------------------------- /problem1/usvs_control.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import model_def 3 | import agent_def 4 | import adp_drl_nn 5 | from math import atan,sin,cos,pi 6 | 7 | NumberofHunters = 3 8 | 9 | Timeoftheworld = 0 10 | 11 | SimulationLimits = 1000000 12 | 13 | center_nums = 72 14 | 15 | allhunters = [None, None, None] 16 | 17 | agent_invader = [None] 18 | 19 | all_critic_1 = [None, None, None] 20 | all_critic_2 = [None, None, None] 21 | all_actor_1 = [None, None, None] 22 | all_actor_2 = [None, None, None] 23 | 24 | z_1_set = [None, None, None] 25 | z_2_set = [None, None, None] 26 | 27 | def create_world(): 28 | 29 | allhunters[0] = model_def.Hunter(0, 1, 7, atan(4), 0, 0, 0) 30 | allhunters[1] = model_def.Hunter(1, 5, 1, atan(1 / 4), 0, 0, 0) 31 | allhunters[2] = model_def.Hunter(2, 2, 8, atan(1 / 3), 0, 0, 0) 32 | 33 | agent_invader[0] = model_def.Invader(0, 8, 6, atan(1.0 / 5)) 34 | 35 | for i in range(NumberofHunters): 36 | 37 | z_1_set[i] = [] 38 | z_2_set[i] = [] 39 | 40 | all_critic_1[i] = adp_drl_nn.Critic1_NN(72, 1) 41 | all_actor_1[i] = adp_drl_nn.Actor1_NN(72, 1) 42 | all_critic_2[i] = adp_drl_nn.Critic2_NN(72, 2) 43 | all_actor_2[i] = adp_drl_nn.Actor2_NN(72, 2) 44 | 45 | def change_state(Timeoftheworld): 46 | 47 | for i in range(NumberofHunters): 48 | # 改变每个智能体的状态 49 | allhunters[i].change_position(allhunters[i].u_hat[0], allhunters[i].u_hat[1]) 50 | # 计算智能体与目标的距离和角度 51 | allhunters[i].get_distance_and_angle(agent_invader[0].pos_x, agent_invader[0].pos_y) 52 | 53 | agent_invader[0].change_speed(0.2, 0.1) 54 | agent_invader[0].change_position() 55 | 56 | def change_network(Timeoftheworld): 57 | 58 | for i in range(NumberofHunters): 59 | 60 | # 得到智能体的邻居 61 | neighbor_id = allhunters[i].get_sametype_neighbors(allhunters) 62 | 63 | # 计算各种系数,并得到z_1 64 | allhunters[i].calculate_super_error(allhunters[neighbor_id[0]].angle, allhunters[neighbor_id[1]].angle, 65 | allhunters[neighbor_id[0]].distance, allhunters[neighbor_id[1]].distance, 66 | allhunters[neighbor_id[0]].speed_u, allhunters[neighbor_id[1]].speed_u, 67 | allhunters[neighbor_id[0]].speed_v, allhunters[neighbor_id[1]].speed_v, 68 | agent_invader[0].speed_u, agent_invader[0].speed_v) 69 | 70 | # RBF网络得到近似值 Step1 71 | z_1_set[i].append(np.norm(allhunters[i].z_1)) 72 | 73 | all_critic_1[i].forward(allhunters[i].z_1) 74 | all_actor_1[i].forward(allhunters[i].z_1) 75 | 76 | allhunters[i].calculate_virtual_opitmal(all_actor_1[i].finaloutputs) 77 | 78 | all_critic_1[i].backward(allhunters[i],all_actor_1[i]) 79 | all_actor_1[i].backward(allhunters[i],all_critic_1[i]) 80 | 81 | # 计算z_2 82 | allhunters[i].calculate_sub_error() 83 | 84 | # RBF网络得到近似值 Step2 85 | z_2_set[i].append(np.norm(allhunters[i].z_2)) 86 | 87 | all_critic_2[i].forward(allhunters[i].z_2) 88 | all_actor_2[i].forward(allhunters[i].z_2) 89 | 90 | allhunters[i].calculate_actual_opitmal(all_actor_2[i].finaloutputs) 91 | 92 | all_critic_2[i].backward(allhunters[i], all_actor_2[i], allhunters[i].optimal_V_hat_dot) 93 | all_actor_2[i].backward(allhunters[i], all_critic_2[i], allhunters[i].optimal_V_hat_dot) 94 | 95 | def train_world(): 96 | 97 | numbercount = 0 98 | 99 | for i in range(SimulationLimits): 100 | 101 | global Timeoftheworld 102 | 103 | change_state(Timeoftheworld) 104 | Timeoftheworld += model_def.T_changeState 105 | 106 | numbercount += 1 107 | 108 | if(numbercount == (model_def.T_changeNN / model_def.T_changeState)): 109 | 110 | change_network(Timeoftheworld) 111 | numbercount = 0 112 | 113 | 114 | if __name__ == "__main__": 115 | 116 | create_world() 117 | train_world() 118 | 119 | --------------------------------------------------------------------------------