Bug fixes from the second full inspection

docs/source/metasim/get_started/advanced_installation/pyroki.md

@@ -9,16 +9,24 @@ PyRoki requires Python 3.10 or higher. Python 3.12+ is recommended for best comp
 ## Installation
 
 ```bash
+cd third_part
 git clone https://github.com/chungmin99/pyroki.git
 cd pyroki
 pip install -e .
+cd ../../
+pip install jax==0.4.30 jaxlib==0.4.30
 ```
-For Isaacsim, also need the following commands:
+
+### For Isaacsim
+If you encounter a NumPy version mismatch between lsaacSim 5.0.0 and PyRoki, for example, an error
+`TypeError: asarray() got an unexpected keyword argument 'copy'`, try running the following commands:
 ```bash
-pip install numpy==1.26.0 # For Isaacsim
-pip install jax==0.6.0 # For Isaacsim
+pip install numpy==1.26.0
+pip install sentry-sdk==1.43.0 typing-extensions==4.12.2 websockets==12.0
+pip install --upgrade websockets
 ```
 
 
 
 
+

docs/source/metasim/get_started/quick_start/4_motion_planning.md

@@ -1,6 +1,6 @@
 # 4. Motion Planning
 In this tutorial, we will show you how to use MetaSim to plan a motion for a robot.
-Note here, we use the `curobo` package to plan the motion. If you haven't installed it, please refer to our [curobo installation guide](https://roboverse.wiki/metasim/get_started/advanced_installation/curobo).
+Note here, we can use the `pyroki` or `curobo` package to plan the motion. If you haven't installed it, please refer to our [pyroki installation guide](https://roboverse.wiki/metasim/get_started/advanced_installation/pyroki) or [curobo installation guide](https://roboverse.wiki/metasim/get_started/advanced_installation/curobo).
 
 ## Common Usage
 

docs/source/metasim/get_started/quick_start/8_replay_demo.md

@@ -9,19 +9,17 @@ python scripts/advanced/replay_demo.py --sim=isaacsim --task=close_box --num_env
 ```
 
 task could also be:
-- `PickCube`
-- `StackCube`
-- `CloseBox`
-- `BasketballInHoop`
+- `pick_cube`
+- `stack_cube`
+- `close_box`
 
 ## States replay
 
 ```bash
 python scripts/advanced/replay_demo.py --sim=isaacsim --task=close_box --num_envs 4 --object-states
 ```
 task could also be:
-- `CloseBox`
-- `BasketballInHoop`
+- `close_box`
 
 ## Varifies commands
 
@@ -30,22 +28,18 @@ task could also be:
 e.g.
 
 ```bash
-python scripts/advanced/replay_demo.py --sim=isaacsim --task=LiberoPickButter
+python scripts/advanced/replay_demo.py --sim=isaacsim --task=libero.pick_butter
 ```
 
 Simulator:
 - `isaacsim`
 - `mujoco`
 
 Task:
-- `LiberoPickAlphabetSoup`
-- `LiberoPickBbqSauce`
-- `LiberoPickChocolatePudding`
-- `LiberoPickCreamCheese`
-- `LiberoPickMilk`
-- `LiberoPickOrangeJuice`
-- `LiberoPickSaladDressing`
-- `LiberoPickTomatoSauce`
+- `libero.kitchen_scene1_open_bottom_drawer`
+- `libero.kitchen_scene1_open_top_drawer`
+- `libero.kitchen_scene1_put_the_black_bowl_on_the_plate`
+
 
 ### Humanoid
 

metasim/sim/hybrid.py

@@ -42,9 +42,9 @@ def close(self) -> None:
         self.physics_handler.close()
         self.render_handler.close()
 
-    def set_dof_targets(self, obj_name: str, actions: list[Action]) -> None:
+    def set_dof_targets(self, actions: list[Action]) -> None:
         """Set the dof targets of the robot in the physics handler."""
-        self.physics_handler.set_dof_targets(obj_name, actions)
+        self.physics_handler.set_dof_targets(actions)
 
     def _set_states(self, states: TensorState, env_ids: list[int] | None = None) -> None:
         """Set states in both physics and render handlers."""

roboverse_learn/il/dp/models/ddpm_image_policy.py

@@ -101,6 +101,10 @@ def conditional_sample(
         # Set diffusion steps.
         scheduler.set_timesteps(self.num_inference_steps)
 
+        # Ensure timesteps are on the same device as trajectory
+        device = trajectory.device
+        scheduler.timesteps = scheduler.timesteps.to(device=device)
+
         step_kwargs = dict(self.scheduler_step_kwargs)
         step_kwargs.update(kwargs)
 
@@ -109,7 +113,7 @@ def conditional_sample(
             trajectory[condition_mask] = condition_data[condition_mask]
 
             # 2. Predict model output.
-            t = t.to(device=trajectory.device)
+           # t = t.to(device=trajectory.device)
             model_output = model(trajectory, t, local_cond=local_cond, global_cond=global_cond)
 
             # 3. Compute previous sample x_t -> x_{t-1}.

roboverse_learn/il/dp/runner/dp_runner.py

@@ -548,7 +548,7 @@ def evaluate(self, ckpt_path=None):
         time_str = datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
         checkpoint = self.get_checkpoint_path()
         # checkpoint = ckpt_path if checkpoint is None else checkpoint
-        checkpoint = ckpt_path if ckpt_path is None else checkpoint
+        checkpoint = ckpt_path if ckpt_path is not None else checkpoint
         if checkpoint is None:
             raise ValueError(
                 "No checkpoint found, please provide a valid checkpoint path."

scripts/advanced/collect_demo.py

@@ -700,20 +700,22 @@ def main():
     task_cls = get_task_class(args.task)
 
     if args.task in {"stack_cube", "pick_cube", "pick_butter"}:
-        dp_camera = True
+        dp_pos = (1.0, 0.0, 0.75)
+    elif args.task in {"close_box"}:
+        dp_pos = (0, 0, 0)
     else:
-        dp_camera = args.task != "close_box"
+        dp_pos = (1.0, 0.0, 0.75)
 
-    is_libero_dataset = "libero_90" in args.task
+    # is_libero_dataset = "libero_90" in args.task
 
-    if is_libero_dataset:
-        dp_pos = (2.0, 0.0, 2)
-    elif dp_camera:
-        # import warnings
-        # warnings.warn("Using dp camera position!")
-        dp_pos = (1.0, 0.0, 0.75)
-    else:
-        dp_pos = (1.5, 0.0, 1.5)
+    # if is_libero_dataset:
+    #     dp_pos = (2.0, 0.0, 2)
+    # elif dp_camera:
+    #     # import warnings
+    #     # warnings.warn("Using dp camera position!")
+    #     dp_pos = (1.0, 0.0, 0.75)
+    # else:
+    #     dp_pos = (1.5, 0.0, 1.5)
 
     # libero specific camera position
     # dp_pos = (0.8, -0, 1.6)
@@ -771,7 +773,7 @@ def main():
     ## Setup
     # Get task description from environment
     task_desc = getattr(env, "task_desc", "")
-    collector = DemoCollector(env.handler, robot, task_desc)
+    collector = DemoCollector(env.handler, robot, task_desc, demo_start_idx=args.demo_start_idx)
     # pbar = tqdm(total=max_demo - args.demo_start_idx, desc="Collecting demos")
     pbar = tqdm(total=args.num_demo_success, desc="Collecting successful demos")
 
@@ -830,13 +832,17 @@ def main():
     stop_flag = False
 
     while not all(finished):
+        if stop_flag:
+            pass
+
         if tot_success >= args.num_demo_success:
-            log.info(f"Reached target number of successful demos ({args.num_demo_success}). Stopping collection.")
-            break
+            log.info(f"Reached target number of successful demos ({args.num_demo_success}).")
+            stop_flag = True
 
         if demo_indexer.next_idx >= max_demo:
-            log.warning(f"Reached maximum demo index ({max_demo}). Stopping collection.")
-            break
+            if not stop_flag:
+                log.warning(f"Reached maximum demo index ({max_demo}), finishing in-flight demos.")
+            stop_flag = True
 
         pbar.set_description(f"Frame {global_step} Success {tot_success} Giveup {tot_give_up}")
         actions = get_actions(all_actions, env, demo_idxs, robot)

scripts/advanced/replay_demo.py

@@ -225,7 +225,7 @@ def main():
             obs = env.handler.get_states()
 
             ## XXX: hack
-            success = env.checker.check(env.handler)
+            success = env.checker.check(env.handler, obs)
             if success.any():
                 log.info(f"Env {success.nonzero().squeeze(-1).tolist()} succeeded!")
             if success.all():