Fixed loss of state (now tested, also with grids), ready to train RL

.gitignore

@@ -47,3 +47,4 @@ roboteam_ai/src/RL/src/tmp/eval_monitor.monitor.csv
 roboteam_ai/src/RL/src/tmp/monitor.monitor.csv
 roboteam_ai/src/RL/src/ppo_tensorboard_main
 roboteam_ai/src/RL/src/best_model_PPO.zip
+docker/runner/ssl-game-controller-config/state-store.json.stream

roboteam_ai/include/roboteam_ai/STPManager.h

@@ -58,22 +58,14 @@ class STPManager {
      */
     void decidePlay(world::World* _world, bool ignoreWorldAge = false);
 
-    /**
-     * @brief Updates the dynamic plays (AttackingPass, Attack, Defend) with the current world state
-     * @details Checks if currentPlay points to a dynamic play and sets it to nullptr if it does.
-     *          Then removes old dynamic plays and creates new ones with updated world information.
-     * @param _world Pointer to the current world state
-     */
-    void updateDynamicPlays(world::World* _world);
-
    public:
     /**
      * @brief Starts the AI with a synchronized boolean to ensure that AI exits correctly
      * @param exitApplication Indicates whether the AI should exit
      */
     void start(std::atomic_flag& exitApplication);
 
-    static PlaysVec plays; /**< The vector that contains all plays */
+    static const PlaysVec plays; /**< The vector that contains all plays */
 
     /**
      * @brief Delete copy constructor of the STPManager class

roboteam_ai/src/STPManager.cpp

@@ -51,10 +51,13 @@ namespace rtt {
 STPManager* STPManager::instance = nullptr;
 
 /// Initialize all plays here (since play vector is static, it's better to do it here to make sure it's initialized before use)
-STPManager::PlaysVec STPManager::plays = ([] {
+const STPManager::PlaysVec STPManager::plays = ([] {
     auto plays = std::vector<std::unique_ptr<ai::stp::Play>>();
 
+    plays.emplace_back(std::make_unique<plays::AttackingPass>());
+    plays.emplace_back(std::make_unique<plays::Attack>());
     plays.emplace_back(std::make_unique<plays::Halt>());
+    plays.emplace_back(std::make_unique<plays::Defend>());
     plays.emplace_back(std::make_unique<plays::KeeperKickBall>());
     plays.emplace_back(std::make_unique<plays::PrepareForcedStart>());
     plays.emplace_back(std::make_unique<plays::StopFormation>());
@@ -73,12 +76,6 @@ STPManager::PlaysVec STPManager::plays = ([] {
     plays.emplace_back(std::make_unique<plays::FreeKickUsPass>());
     plays.emplace_back(std::make_unique<plays::KickOffUs>());
     plays.emplace_back(std::make_unique<plays::KickOffThem>());
-
-    // Dynamic plays (these will be recreated each tick)
-    plays.emplace_back(std::make_unique<plays::AttackingPass>());
-    plays.emplace_back(std::make_unique<plays::Attack>());
-    plays.emplace_back(std::make_unique<plays::Defend>());
-
     return plays;
 })();
 
@@ -194,10 +191,6 @@ void STPManager::decidePlay(world::World *_world, bool ignoreWorldAge) {
             return;
         }
     }
-
-    // Update the dynamic plays (AttackingPass, Attack, Defend)
-    STPManager::updateDynamicPlays(_world);
-
     if (!currentPlay || !currentPlay->isValidPlayToKeep() || ai::stp::PlayDecider::didLockPlay()) {
         // Decide the best play (ignoring the interface play value)
         currentPlay = ai::stp::PlayDecider::decideBestPlay(_world, plays);
@@ -209,38 +202,6 @@ void STPManager::decidePlay(world::World *_world, bool ignoreWorldAge) {
     currentPlay->update();
 }
 
-void STPManager::updateDynamicPlays(world::World* _world) {
-
-    // Create the new plays (attackingPass, attack, defend) and set their world immediately
-    auto attackingPass = std::make_unique<plays::AttackingPass>();
-    auto attack = std::make_unique<plays::Attack>();
-    auto defend = std::make_unique<plays::Defend>();
-
-    attackingPass->setWorld(_world);
-    attack->setWorld(_world);
-    defend->setWorld(_world);
-
-    // Check if currentPlay is one of the dynamic plays that we're about to delete
-    if (currentPlay) {
-        for (int i = plays.size() - 3; i < plays.size(); ++i) {
-            if (currentPlay == plays[i].get()) {
-                currentPlay = nullptr; // If it is, set to nullptr. Otherwise currentPlay will point to deleted memory
-                break;
-            }
-        }
-    }
-
-    // Remove old plays
-    for (int i = 0; i < 3; i++) {
-        plays.pop_back();
-    }
-
-    // Add new plays
-    plays.emplace_back(std::move(attackingPass));
-    plays.emplace_back(std::move(attack));
-    plays.emplace_back(std::move(defend));
-}
-
 STPManager::STPManager(std::shared_ptr<ai::gui::net::InterfaceGateway> interfaceGateway) : 
     interfaceGateway(std::move(interfaceGateway)),
     rlInterface() {
@@ -249,7 +210,7 @@ STPManager::STPManager(std::shared_ptr<ai::gui::net::InterfaceGateway> interface
         // Set initial grid positions
         std::array<bool, 9> defaultGrid = {
             false,  false, true,   // top row
-            false, true, true,  // middle row
+            false, false, false,  // middle row
             false,  false, true    // bottom row
         };
         rlInterface.setBinaryOccupancyGrid(defaultGrid);

roboteam_ai/src/rl/zmq_PUB.py

@@ -21,7 +21,7 @@ def create_publisher():
 
 def publish_messages(socket, topic="updates"):
     # Array of integers to send
-    numbers = [3, 7, 9, 8]
+    numbers = [2, 3,6]
 
     while True:
         # Convert numbers to strings

roboteam_ai/src/stp/plays/offensive/AttackingPass.cpp

@@ -88,7 +88,7 @@ uint8_t AttackingPass::score(const rtt::Field& field) noexcept {
 Dealer::FlagMap AttackingPass::decideRoleFlags() const noexcept {
     int numDefenders = 2;
     int numWallers = 4;
-    int numAttackers = 2;
+    int numAttackers = STPManager::getRLInterface().getNumAttackers();
 
     Dealer::FlagMap flagMap;