test passed

envpool/minigrid/BUILD

@@ -69,5 +69,6 @@ py_test(
         requirement("dm_env"),
         requirement("gym"),
         requirement("numpy"),
+        requirement("minigrid"),
     ],
 )

envpool/minigrid/impl/minigrid_env.cc

@@ -19,6 +19,7 @@ namespace minigrid {
 void MiniGridEnv::MiniGridReset() {
   GenGrid();
   step_count_ = 0;
+  done_ = false;
   CHECK(agent_pos_.first >= 0 && agent_pos_.second >= 0);
   CHECK(agent_dir_ >= 0);
   CHECK(grid_[agent_pos_.second][agent_pos_.first].GetType() == kEmpty);
@@ -54,8 +55,9 @@ float MiniGridEnv::MiniGridStep(Act act) {
   // Get the forward cell object
   if (act == kLeft) {
     agent_dir_ -= 1;
-    if (agent_dir_ < 0)
+    if (agent_dir_ < 0) {
       agent_dir_ += 4;
+    }
   } else if (act == kRight) {
     agent_dir_ = (agent_dir_ + 1) % 4;
   } else if (act == kForward) {
@@ -182,8 +184,11 @@ void MiniGridEnv::GenImage(Array& obs) {
     memset(vis_mask, 1, sizeof(vis_mask));
   }
   // Let the agent see what it's carrying
-  if (carrying_.GetType() != kEmpty)
+  if (carrying_.GetType() != kEmpty) {
     agent_view_grid[agent_pos_y][agent_pos_x] = carrying_;
+  } else {
+    agent_view_grid[agent_pos_y][agent_pos_x] = WorldObj(kEmpty);
+  }
   for (int y = 0; y < agent_view_size_; ++y) {
     for (int x = 0; x < agent_view_size_; ++x) {
       if (vis_mask[y * agent_view_size_ + x] == true) {

envpool/minigrid/minigrid_test.py

@@ -15,8 +15,11 @@
 
 from typing import Any
 
+import gymnasium as gym
+import minigrid
 import numpy as np
 from absl.testing import absltest
+from absl import logging
 
 import envpool.minigrid.registration  # noqa: F401
 from envpool.registration import make_gym
@@ -28,11 +31,35 @@ def test_deterministic_check(
     self,
     task_id: str = "MiniGrid-Empty-5x5-v0",
     num_envs: int = 1,
+    total: int = 100000,
     **kwargs: Any,
   ) -> None:
-    env = make_gym(task_id, num_envs=num_envs, seed=0, **kwargs)
-    obs, info = env.reset()
-    print(obs)
+    env0 = gym.make(task_id)
+    env1 = make_gym(task_id, num_envs=num_envs, seed=0, **kwargs)
+    obs0, info0 = env0.reset()
+    obs1, info1 = env1.reset()
+    np.testing.assert_allclose(obs0["image"], obs1["image"][0])
+    done0 = False
+    acts = []
+    for i in range(total):
+      act = env0.action_space.sample()
+      acts.append(act)
+      if done0:
+        obs0, info0 = env0.reset()
+        auto_reset = True
+        term0 = trunc0 = False
+      else:
+        obs0, rew0, term0, trunc0, info0 = env0.step(act)
+        auto_reset = False
+      obs1, rew1, term1, trunc1, info1 = env1.step(np.array([act]))
+      self.assertEqual(obs0["image"].shape, (7, 7, 3))
+      self.assertEqual(obs1["image"].shape, (num_envs, 7, 7, 3))
+      done0 = term0 | trunc0
+      done1 = term1 | trunc1
+      if not auto_reset:
+        np.testing.assert_allclose(rew0, rew1[0], rtol=1e-6)
+        np.testing.assert_allclose(done0, done1[0])
+      np.testing.assert_allclose(obs0["image"], obs1["image"][0])
 
 
 if __name__ == "__main__":

third_party/pip_requirements/requirements-dev.txt

@@ -18,3 +18,4 @@ dm-control>=1.0.5
 mujoco>=2.2.1,<2.3
 mujoco_py>=2.1.2.14
 pygame
+minigrid