I am trying to make an RL model for reinforcement learning, and the environment is fairly simple and a 10×10 grid. During the training phase, it reaches the goal. But when I try to test this, The agent takes one direction and goes straight and hits a wall or obstacle. It won’t change direction. And this goes on in a loop. My reward function is based on distance from the current position to the goal and penalties are given for moving away from the goal and hitting anything. The following given is my model class.

class DQNTrainer:
    def __init__(self, states, actions, visualize=True, load_model=False):
        self.states = states
        self.actions = actions
        self.model = self.create_model()
        if load_model:
            self.load_model()
        self.agent = self.create_agent()
        self.env = None
        self.visualize = visualize

    def create_model(self):
        model = Sequential()
        model.add(Flatten(input_shape=(1, self.states, self.states)))
        model.add(Dense(128, activation="relu", kernel_regularizer=l1(0.01)))  # Add L1 regularization
        model.add(Dense(128, activation="relu", kernel_regularizer=l1(0.01)))
        model.add(Dense(64, activation="relu", kernel_regularizer=l1(0.01)))
        model.add(Dense(self.actions, activation="linear"))
        return model

    def create_agent(self):
        policy = LinearAnnealedPolicy(EpsGreedyQPolicy(), 
                                  attr='eps', 
                                  value_max=1.0, 
                                  value_min=0.1, 
                                  value_test=0.01, 
                                  nb_steps=10000)
        return DQNAgent(
            model=self.model,
            memory=SequentialMemory(limit=20000, window_length=1),
            policy=policy,
            #policy = EpsGreedyQPolicy(0.1),
            nb_actions=self.actions,
            nb_steps_warmup=500,
            target_model_update=0.005,
            gamma=0.99
        )

    def compile_agent(self):
        self.agent.compile(optimizer=Adam(learning_rate=0.0005), 
                           metrics=['mean_absolute_error'])

    def fit(self, env, nb_steps):
        self.env = env
        self.agent.fit(self.env, nb_steps=nb_steps, visualize=self.visualize, verbose=1)

    def train(self, env, nb_steps):
        self.compile_agent()
        self.fit(env, nb_steps)
        return True

    def save_model(self):
        self.model.save_weights('pathplanner.h5', overwrite=True)

    def load_model(self):
        try:
            self.model.load_weights('pathplanner.h5')
            print("Model Loading Successful")
        except Exception as e:
            print(f"Error loading model: {e}")