Upload folder using huggingface_hub

models.py

@@ -6,6 +6,7 @@ class MazeAction(Action):
 
 class MazeObservation(Observation):
     position: list = Field(default=[], description="Agent's [row, col]")
+    valid_moves: list = Field(default=[], description="List of valid directions")
     grid_view: str = Field(default="", description="String view of the maze")
 
 class MazeState(State):

server/maze_environment.py

@@ -53,15 +53,35 @@ class MazeEnvironment(Environment):
         if seed: random.seed(seed)
         self._generate_new_maze()
         self._episode_id = episode_id
-        return MazeObservation(position=self._agent_pos, grid_view=self._render(), done=False, reward=0)
+        return MazeObservation(
+            position=self._agent_pos,
+            grid_view=self._render(),
+            valid_moves=self._get_valid_moves(),  # add this
+            done=False,
+            reward=0
+        )
+
+    def _get_valid_moves(self) -> list:
+        moves = []
+        i, j = self._agent_pos
+        if i > 0 and self._maze[i-1][j]: moves.append("up")
+        if i < self.row-1 and self._maze[i+1][j]: moves.append("down")
+        if j > 0 and self._maze[i][j-1]: moves.append("left")
+        if j < self.col-1 and self._maze[i][j+1]: moves.append("right")
+        return moves
 
     def step(self, action: MazeAction, timeout_s=None, **kwargs) -> MazeObservation:
         if action.direction in ["up", "down", "left", "right"]:
             self._move(action.direction)
         self._step_count += 1
         done = self._agent_pos == self._goal_pos
-        return MazeObservation(position=self._agent_pos, grid_view=self._render(),
-                               done=done, reward=10 if done else -1)
+        return MazeObservation(
+            position=self._agent_pos,
+            grid_view=self._render(),
+            valid_moves=self._get_valid_moves(),  # add this
+            done=done,
+            reward=10 if done else -1
+        )
 
     def _move(self, d):
         i, j = self._agent_pos