Fixes warnings when running AutoMate env

source/isaaclab_tasks/isaaclab_tasks/direct/automate/assembly_env.py

@@ -66,19 +66,6 @@ def __init__(self, cfg: AssemblyEnvCfg, render_mode: str | None = None, **kwargs
         if self.cfg_task.if_logging_eval:
             self._init_eval_logging()
 
-        if self.cfg_task.sample_from != "rand":
-            self._init_eval_loading()
-
-    def _init_eval_loading(self):
-        eval_held_asset_pose, eval_fixed_asset_pose, eval_success = automate_log.load_log_from_hdf5(
-            self.cfg_task.eval_filename
-        )
-
-        if self.cfg_task.sample_from == "gp":
-            self.gp = automate_algo.model_succ_w_gp(eval_held_asset_pose, eval_fixed_asset_pose, eval_success)
-        elif self.cfg_task.sample_from == "gmm":
-            self.gmm = automate_algo.model_succ_w_gmm(eval_held_asset_pose, eval_fixed_asset_pose, eval_success)
-
     def _init_eval_logging(self):
 
         self.held_asset_pose_log = torch.empty(
@@ -246,7 +233,7 @@ def _load_disassembly_data(self):
             # offset each trajectory to be relative to the goal
             eef_pos_traj.append(curr_ee_traj - curr_ee_goal)
 
-        self.eef_pos_traj = torch.tensor(eef_pos_traj, dtype=torch.float32, device=self.device).squeeze()
+        self.eef_pos_traj = torch.tensor(np.array(eef_pos_traj), dtype=torch.float32, device=self.device).squeeze()
 
     def _get_keypoint_offsets(self, num_keypoints):
         """Get uniformly-spaced keypoints along a line of unit length, centered at 0."""
@@ -804,28 +791,12 @@ def randomize_held_initial_state(self, env_ids, pre_grasp):
                 torch.rand((self.num_envs,), dtype=torch.float32, device=self.device)
             )
 
-            if self.cfg_task.sample_from == "rand":
-
-                rand_sample = torch.rand((len(env_ids), 3), dtype=torch.float32, device=self.device)
-                held_pos_init_rand = 2 * (rand_sample - 0.5)  # [-1, 1]
-                held_asset_init_pos_rand = torch.tensor(
-                    self.cfg_task.held_asset_init_pos_noise, dtype=torch.float32, device=self.device
-                )
-                self.held_pos_init_rand = held_pos_init_rand @ torch.diag(held_asset_init_pos_rand)
-
-            if self.cfg_task.sample_from == "gp":
-                rand_sample = torch.rand((self.cfg_task.num_gp_candidates, 3), dtype=torch.float32, device=self.device)
-                held_pos_init_rand = 2 * (rand_sample - 0.5)  # [-1, 1]
-                held_asset_init_pos_rand = torch.tensor(
-                    self.cfg_task.held_asset_init_pos_noise, dtype=torch.float32, device=self.device
-                )
-                held_asset_init_candidates = held_pos_init_rand @ torch.diag(held_asset_init_pos_rand)
-                self.held_pos_init_rand, _ = automate_algo.propose_failure_samples_batch_from_gp(
-                    self.gp, held_asset_init_candidates.cpu().detach().numpy(), len(env_ids), self.device
-                )
-
-            if self.cfg_task.sample_from == "gmm":
-                self.held_pos_init_rand = automate_algo.sample_rel_pos_from_gmm(self.gmm, len(env_ids), self.device)
+            rand_sample = torch.rand((len(env_ids), 3), dtype=torch.float32, device=self.device)
+            held_pos_init_rand = 2 * (rand_sample - 0.5)  # [-1, 1]
+            held_asset_init_pos_rand = torch.tensor(
+                self.cfg_task.held_asset_init_pos_noise, dtype=torch.float32, device=self.device
+            )
+            self.held_pos_init_rand = held_pos_init_rand @ torch.diag(held_asset_init_pos_rand)
 
         # Set plug pos to assembled state, but offset plug Z-coordinate by height of socket,
         # minus curriculum displacement

source/isaaclab_tasks/isaaclab_tasks/direct/automate/assembly_tasks_cfg.py

@@ -139,10 +139,6 @@ class AssemblyTask:
     num_eval_trials: int = 100
     eval_filename: str = "evaluation_00015.h5"
 
-    # Fine-tuning
-    sample_from: str = "rand"  # gp, gmm, idv, rand
-    num_gp_candidates: int = 1000
-
 
 @configclass
 class Peg8mm(HeldAssetCfg):

source/isaaclab_tasks/isaaclab_tasks/direct/automate/disassembly_tasks_cfg.py

@@ -118,7 +118,7 @@ class Extraction(DisassemblyTask):
     assembly_id = "00015"
     assembly_dir = f"{ASSET_DIR}/{assembly_id}/"
     disassembly_dir = "disassembly_dir"
-    num_log_traj = 1000
+    num_log_traj = 100
 
     fixed_asset_cfg = Hole8mm()
     held_asset_cfg = Peg8mm()

source/isaaclab_tasks/isaaclab_tasks/direct/automate/run_w_id.py

@@ -61,9 +61,9 @@ def main():
 
     bash_command = None
     if sys.platform.startswith("win"):
-        bash_command = "isaaclab.bat -p"
+        bash_command = "NUMBA_CUDA_LOW_OCCUPANCY_WARNINGS=0 isaaclab.bat -p"
     elif sys.platform.startswith("linux"):
-        bash_command = "./isaaclab.sh -p"
+        bash_command = "NUMBA_CUDA_LOW_OCCUPANCY_WARNINGS=0 ./isaaclab.sh -p"
     if args.train:
         bash_command += " scripts/reinforcement_learning/rl_games/train.py --task=Isaac-AutoMate-Assembly-Direct-v0"
         bash_command += f" --seed={str(args.seed)} --max_iterations={str(args.max_iterations)}"

source/isaaclab_tasks/isaaclab_tasks/direct/automate/soft_dtw_cuda.py

@@ -123,8 +123,8 @@ class _SoftDTWCUDA(Function):
     def forward(ctx, D, gamma, bandwidth):
         dev = D.device
         dtype = D.dtype
-        gamma = torch.cuda.FloatTensor([gamma])
-        bandwidth = torch.cuda.FloatTensor([bandwidth])
+        gamma = torch.tensor([gamma], dtype=torch.float, device="cuda")
+        bandwidth = torch.tensor([bandwidth], dtype=torch.float, device="cuda")
 
         B = D.shape[0]
         N = D.shape[1]