BUGFIX: Fix object-target proximity threshold

docs/source/challenge-doc.rst

@@ -174,7 +174,7 @@ Success Condition
 
     - The object moved from start position to goal position. Both the MPL hand, and MyoHand, is required to touch the object for certain timesteps 
     - Exerting a maximum contact force on the object, less than 1500N (subject to change in final EVALUATION environment)
-    - Placing the object within 0.05 meters of the goal site on the pillar
+    - Placing the object within 0.17 meters of the goal site on the pillar (0.1m in each of xyz-axis)
 
 
 Ranking Criteria

myosuite/agents/sb3_job_script.py

@@ -62,6 +62,7 @@ def train_loop(job_data) -> None:
                     learning_rate=job_data.learning_rate, 
                     batch_size=job_data.batch_size, 
                     policy_kwargs=policy_kwargs,
+                    tensorboard_log=f"wandb/{run.id}",
                     gamma=job_data.gamma, **job_data.alg_hyper_params)
     elif algo == 'SAC':
         model = SAC(job_data.policy, env, 
@@ -70,6 +71,7 @@ def train_loop(job_data) -> None:
                     learning_starts=job_data.learning_starts, 
                     batch_size=job_data.batch_size, 
                     tau=job_data.tau, 
+                    tensorboard_log=f"wandb/{run.id}",
                     gamma=job_data.gamma, **job_data.alg_hyper_params)
 
 

myosuite/envs/myo/myochallenge/bimanual_v0.py

@@ -50,7 +50,7 @@ def _setup(self,
                goal_center=np.array([0.4, -0.25, 1.05]),
                max_force=1500,  # Max force against throwing
 
-               proximity_th=0.015,  # object-target proximity threshold
+               proximity_th=0.17,  # object-target proximity threshold, based on 10cm in each axis in Euclidean Distance
 
                start_shifts=np.array([0.055, 0.055, 0]),
                # shift factor for start/goal random generation with z-axis fixed
@@ -78,6 +78,7 @@ def _setup(self,
 
         self.start_shifts = start_shifts
         self.goal_shifts = goal_shifts
+        self.PILLAR_HEIGHT = 1.09
 
         self.id_info = IdInfo(self.sim.model)
 
@@ -246,6 +247,7 @@ def get_reward_dict(self, obs_dict):
         obj_pos = obs_dict["obj_pos"][0][0] if obs_dict['obj_pos'].ndim == 3 else obs_dict['obj_pos']
         palm_pos = obs_dict["palm_pos"][0][0] if obs_dict["palm_pos"].ndim == 3 else obs_dict["palm_pos"]
         goal_pos = obs_dict["goal_pos"][0][0] if obs_dict["goal_pos"].ndim == 3 else obs_dict["goal_pos"]
+        goal_pos = np.concatenate((goal_pos[:2], np.array([self.PILLAR_HEIGHT])))
 
         lift_height = np.linalg.norm(np.array([[[obj_pos[-1], palm_pos[-1]]]]) -
                                      np.array([[[self.init_obj_z, self.init_palm_z]]]), axis=-1)