fix the setup

sim/envs/humanoids/stompypro_config.py

@@ -151,7 +151,7 @@ class rewards:
         # put some settings here for LLM parameter tuning
         target_joint_pos_scale = 0.17  # rad
         target_feet_height = 0.05  # m
-        cycle_time = 0.4  # sec
+        cycle_time = 0.4  # sec # pfb30 - that should be longer and more stable
         # if true negative total rewards are clipped at zero (avoids early termination problems)
         only_positive_rewards = True
         # tracking reward = exp(error*sigma)

sim/resources/stompypro/joints.py

@@ -180,6 +180,16 @@ def stiffness(cls) -> Dict[str, float]:
             "knee": 150,
             "ankle_y": 75,
         }
+
+    @classmethod
+    def stiffness_mujoco(cls) -> Dict[str, float]:
+        return {
+            "hip_y": 120,
+            "hip_x": 60,
+            "hip_z": 60,
+            "knee": 120,
+            "ankle_y": 17,
+        }
 
     # d_gains
     @classmethod
@@ -192,24 +202,35 @@ def damping(cls) -> Dict[str, float]:
             "ankle_y": 5,
         }
 
-    # pos_limits
-    @classmethod
-    def effort(cls) -> Dict[str, float]:
-        return {}
 
-    # vel_limits
     @classmethod
-    def velocity(cls) -> Dict[str, float]:
+    def damping_mujoco(cls) -> Dict[str, float]:
         return {
-            "hip": 5,
+            "hip_y": 5,
+            "hip_x": 5,
+            "hip_z": 5,
             "knee": 5,
-            "ankle": 5,
-            "shoulder_y": 5,
-            "shoulder_z": 5,
-            "shoulder_x": 5,
-            "elbow_x": 5,
+            "ankle_y": 5,
         }
 
+    # # pos_limits
+    # @classmethod
+    # def effort(cls) -> Dict[str, float]:
+    #     return {}
+
+    # # vel_limits
+    # @classmethod
+    # def velocity(cls) -> Dict[str, float]:
+    #     return {
+    #         "hip": 5,
+    #         "knee": 5,
+    #         "ankle": 5,
+    #         "shoulder_y": 5,
+    #         "shoulder_z": 5,
+    #         "shoulder_x": 5,
+    #         "elbow_x": 5,
+    #     }
+
     @classmethod
     def friction(cls) -> Dict[str, float]:
         return {

sim/sim2sim.py

@@ -151,8 +151,8 @@ def run_mujoco(policy, cfg):
 
             cur_vel_obs = dq * cfg.normalization.obs_scales.dof_vel
 
-            obs[0, 0] = math.sin(2 * math.pi * count_lowlevel * cfg.sim_config.dt / 0.64)
-            obs[0, 1] = math.cos(2 * math.pi * count_lowlevel * cfg.sim_config.dt / 0.64)
+            obs[0, 0] = math.sin(2 * math.pi * count_lowlevel * cfg.sim_config.dt / 0.4)
+            obs[0, 1] = math.cos(2 * math.pi * count_lowlevel * cfg.sim_config.dt / 0.4)
             obs[0, 2] = cmd.vx * cfg.normalization.obs_scales.lin_vel
             obs[0, 3] = cmd.vy * cfg.normalization.obs_scales.lin_vel
             obs[0, 4] = cmd.dyaw * cfg.normalization.obs_scales.ang_vel
@@ -184,6 +184,7 @@ def run_mujoco(policy, cfg):
 
         tau = np.clip(tau, -cfg.robot_config.tau_limit, cfg.robot_config.tau_limit)  # Clamp torques
         print(tau)
+        # breakpoint()
         data.ctrl = tau
 
         mujoco.mj_step(model, data)
@@ -215,13 +216,17 @@ class env:
 
         class sim_config:
             sim_duration = 60.0
-            dt = 0.001
-            decimation = 20
+            dt = 0.002
+            decimation = 10
+
+        class rewards:
+            cycle_time = 0.4
 
         class robot_config:
             tau_factor = 0.85
-            tau_limit = np.array(list(robot.stiffness().values()) + list(robot.stiffness().values())) * tau_factor
-            kps = tau_limit
+            # pfb30 - bring back the original setup but fix the logic or call effort to have the same logic as in the script
+            tau_limit = np.array(list(robot.stiffness_mujoco().values()) + list(robot.stiffness_mujoco().values())) * tau_factor
+            kps = np.array(list(robot.stiffness().values()) + list(robot.stiffness().values()))
             kds = np.array(list(robot.damping().values()) + list(robot.damping().values()))
 
         class normalization: