gpr - smaller efforts, updates to export

sim/envs/__init__.py

@@ -12,8 +12,8 @@
 from sim.envs.humanoids.dora_env import DoraFreeEnv
 from sim.envs.humanoids.g1_config import G1Cfg, G1CfgPPO
 from sim.envs.humanoids.g1_env import G1FreeEnv
-from sim.envs.humanoids.gpr_env import GprFreeEnv
 from sim.envs.humanoids.gpr_config import GprCfg, GprCfgPPO, GprStandingCfg
+from sim.envs.humanoids.gpr_env import GprFreeEnv
 from sim.envs.humanoids.h1_config import H1Cfg, H1CfgPPO
 from sim.envs.humanoids.h1_env import H1FreeEnv
 from sim.envs.humanoids.xbot_config import XBotCfg, XBotCfgPPO

sim/envs/humanoids/gpr_config.py

@@ -30,7 +30,7 @@ class safety(LeggedRobotCfg.safety):
         # safety factors
         pos_limit = 1.0
         vel_limit = 1.0
-        torque_limit = 0.85
+        torque_limit = 1.0
 
     class asset(LeggedRobotCfg.asset):
         name = "gpr"
@@ -123,7 +123,7 @@ class domain_rand(LeggedRobotCfg.domain_rand):
 
         randomize_base_mass = True
         added_mass_range = [-3.0, 3.0]
-        push_robots = False
+        push_robots = True
         push_interval_s = 4
         max_push_vel_xy = 0.2
         max_push_ang_vel = 0.4
@@ -150,14 +150,14 @@ class rewards:
         max_dist = 0.5
         # put some settings here for LLM parameter tuning
         target_joint_pos_scale = 0.17  # rad
-        target_feet_height = 0.06  # m
+        target_feet_height = 0.05  # m
 
-        cycle_time = 0.64  # sec
+        cycle_time = 0.25  # sec
         # if true negative total rewards are clipped at zero (avoids early termination problems)
         only_positive_rewards = True
         # tracking reward = exp(error*sigma)
         tracking_sigma = 5.0
-        max_contact_force = 700  # forces above this value are penalized
+        max_contact_force = 400  # forces above this value are penalized
 
         class scales:
             # reference motion tracking
@@ -213,7 +213,7 @@ class GprStandingCfg(GprCfg):
 
     class rewards:
         # quite important to keep it right
-        base_height_target = 1.06
+        base_height_target = Robot.height
         min_dist = 0.2
         max_dist = 0.5
         # put some settings here for LLM parameter tuning

sim/model_export.py

@@ -15,17 +15,20 @@
 
 @dataclass
 class ActorCfg:
-    embodiment: str = "gpr"
-    cycle_time: float = 0.4  # Cycle time for sinusoidal command input
-    action_scale: float = 0.25  # Scale for actions
-    lin_vel_scale: float = 2.0  # Scale for linear velocity
-    ang_vel_scale: float = 1.0  # Scale for angular velocity
-    quat_scale: float = 1.0  # Scale for quaternion
-    dof_pos_scale: float = 1.0  # Scale for joint positions
-    dof_vel_scale: float = 0.05  # Scale for joint velocities
-    frame_stack: int = 15  # Number of frames to stack for the policy input
-    clip_observations: float = 18.0  # Clip observations to this value
-    clip_actions: float = 18.0  # Clip actions to this value
+    embodiment: str
+    cycle_time: float  # Cycle time for sinusoidal command input
+    action_scale: float  # Scale for actions
+    lin_vel_scale: float  # Scale for linear velocity
+    ang_vel_scale: float  # Scale for angular velocity
+    quat_scale: float  # Scale for quaternion
+    dof_pos_scale: float  # Scale for joint positions
+    dof_vel_scale: float  # Scale for joint velocities
+    frame_stack: int  # Number of frames to stack for the policy input
+    clip_observations: float  # Clip observations to this value
+    clip_actions: float  # Clip actions to this value
+    sim_dt: float  # Simulation time step
+    sim_decimation: int  # Simulation decimation
+    tau_factor: float  # Torque limit factor
 
 
 class ActorCritic(nn.Module):
@@ -242,6 +245,8 @@ def get_actor_policy(model_path: str, cfg: ActorCfg) -> Tuple[nn.Module, dict, T
     num_actions = a_model.num_actions
     num_observations = a_model.num_observations
 
+    default_standing = list(a_model.robot.default_standing().values())
+
     return (
         a_model,
         {
@@ -250,6 +255,7 @@ def get_actor_policy(model_path: str, cfg: ActorCfg) -> Tuple[nn.Module, dict, T
             "robot_damping": robot_damping,
             "num_actions": num_actions,
             "num_observations": num_observations,
+            "default_standing": default_standing,
         },
         input_tensors,
     )
@@ -338,7 +344,3 @@ def convert_model_to_onnx(model_path: str, cfg: ActorCfg, save_path: Optional[st
     buffer2.seek(0)
 
     return ort.InferenceSession(buffer2.read())
-
-
-if __name__ == "__main__":
-    convert_model_to_onnx("model_3000.pt", ActorCfg(), "policy.onnx")

sim/play.py

@@ -21,15 +21,11 @@
 from kinfer.export.pytorch import export_to_onnx
 from tqdm import tqdm
 
-# Local imports third
 from sim.env import run_dir  # noqa: E402
 from sim.envs import task_registry  # noqa: E402
-from sim.h5_logger import HDF5Logger
-from sim.model_export import (  # noqa: E402
-    ActorCfg,
-    convert_model_to_onnx,
-    get_actor_policy,
-)
+
+# Local imports third
+from sim.model_export import ActorCfg, get_actor_policy
 from sim.utils.helpers import get_args  # noqa: E402
 from sim.utils.logger import Logger  # noqa: E402
 
@@ -92,15 +88,22 @@ def play(args: argparse.Namespace) -> None:
     if args.export_onnx:
         path = ppo_runner.load_path
         embodiment = ppo_runner.cfg["experiment_name"].lower()
-        policy_cfg = ActorCfg(embodiment=embodiment)
-
-        if embodiment == "stompypro":
-            policy_cfg.cycle_time = 0.4
-        elif embodiment == "stompymicro":
-            policy_cfg.cycle_time = 0.2
-        else:
-            print(f"Specific policy cfg for {embodiment} not implemented")
-
+        policy_cfg = ActorCfg(
+            embodiment=embodiment,
+            cycle_time=env_cfg.rewards.cycle_time,
+            sim_dt=env_cfg.sim.dt,
+            sim_decimation=env_cfg.control.decimation,
+            tau_factor=env_cfg.safety.torque_limit,
+            action_scale=env_cfg.control.action_scale,
+            lin_vel_scale=env_cfg.normalization.obs_scales.lin_vel,
+            ang_vel_scale=env_cfg.normalization.obs_scales.ang_vel,
+            quat_scale=env_cfg.normalization.obs_scales.quat,
+            dof_pos_scale=env_cfg.normalization.obs_scales.dof_pos,
+            dof_vel_scale=env_cfg.normalization.obs_scales.dof_vel,
+            frame_stack=env_cfg.env.frame_stack,
+            clip_observations=env_cfg.normalization.clip_observations,
+            clip_actions=env_cfg.normalization.clip_actions,
+        )
         actor_model, sim2sim_info, input_tensors = get_actor_policy(path, policy_cfg)
 
         # Merge policy_cfg and sim2sim_info into a single config object
@@ -117,6 +120,8 @@ def play(args: argparse.Namespace) -> None:
 
     now = datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
     if args.log_h5:
+        from sim.h5_logger import HDF5Logger
+
         # Create directory for HDF5 files
         h5_dir = run_dir() / "h5_out" / args.task / now
         h5_dir.mkdir(parents=True, exist_ok=True)