add more logging

sim/h5_logger.py

@@ -24,7 +24,7 @@ def _create_h5_file(self):
         h5_file = h5py.File(f"{self.data_name}/{idd}.h5", "w")
 
         # Create datasets for logging actions and observations
-        dset_actions = h5_file.create_dataset("actions", (self.max_timesteps, self.num_actions), dtype=np.float32)
+        dset_actions = h5_file.create_dataset("prev_actions", (self.max_timesteps, self.num_actions), dtype=np.float32)
         dset_2D_command = h5_file.create_dataset("observations/2D_command", (self.max_timesteps, 2), dtype=np.float32)
         dset_3D_command = h5_file.create_dataset("observations/3D_command", (self.max_timesteps, 3), dtype=np.float32)
         dset_q = h5_file.create_dataset("observations/q", (self.max_timesteps, self.num_actions), dtype=np.float32)
@@ -36,7 +36,7 @@ def _create_h5_file(self):
 
         # Map datasets for easy access
         h5_dict = {
-            "actions": dset_actions,
+            "prev_actions": dset_actions,
             "2D_command": dset_2D_command,
             "3D_command": dset_3D_command,
             "joint_pos": dset_q,
@@ -123,4 +123,4 @@ def _plot_dataset(name: str, data: np.ndarray):
 
 
 if __name__ == "__main__":
-    HDF5Logger.visualize_h5("stompypro/6dc85e02-fc8e-42e1-a396-b0bd578e0816.h5")
+    HDF5Logger.visualize_h5("stompypro/5a7dc371-445c-4f56-be05-4e65c5cc38bc.h5")

sim/produce_sim_data.py

@@ -1,4 +1,4 @@
-"""Produce simulation data for training."""
+""" Produce simulation data for training."""
 import argparse
 import multiprocessing as mp
 import subprocess
@@ -47,7 +47,7 @@ def run_parallel_sims(num_threads: int, args: argparse.Namespace) -> None:
     parser.add_argument("--num_threads", type=int, default=10, help="Number of parallel simulations to run")
     parser.add_argument("--embodiment", default="stompypro", type=str, help="Embodiment name")
     parser.add_argument("--load_model", default="examples/walking_pro.onnx", type=str, help="Path to model to load")
-    parser.add_argument("--num_examples", default=1000, type=int, help="Number of examples to run")
+    parser.add_argument("--num_examples", default=10000, type=int, help="Number of examples to run")
     args = parser.parse_args()
 
     # Run 100 examples total, in parallel batches

sim/sim2sim.py

@@ -7,12 +7,10 @@
 import argparse
 import math
 import os
-import uuid
 from copy import deepcopy
 from dataclasses import dataclass
 from typing import Dict, List, Tuple, Union
 
-import h5py
 import mujoco
 import mujoco_viewer
 import numpy as np
@@ -31,6 +29,8 @@ class Sim2simCfg:
     dt: float = 0.001
     decimation: int = 10
     tau_factor: float = 3
+    cycle_time: float = 0.4
+
 
 
 def handle_keyboard_input() -> None:
@@ -185,7 +185,6 @@ def run_mujoco(
     total_speed = 0.0
     step_count = 0
 
-    t = 0
     for _ in tqdm(range(int(cfg.sim_duration / cfg.dt)), desc="Simulating..."):
         if keyboard_use:
             handle_keyboard_input()
@@ -232,22 +231,28 @@ def run_mujoco(
 
             input_data["buffer.1"] = hist_obs.astype(np.float32)
 
-            positions, actions, hist_obs = policy.run(None, input_data)
-            target_q = positions
-
             if args.log_h5:
                 logger.log_data({
                     "t": np.array([count_lowlevel * cfg.dt], dtype=np.float32),
-                    "2D_command": np.array([x_vel_cmd, y_vel_cmd], dtype=np.float32),
+                    "2D_command": np.array(
+                        [
+                            np.sin(2 * math.pi * count_lowlevel * cfg.dt / cfg.cycle_time),
+                            np.cos(2 * math.pi * count_lowlevel * cfg.dt / cfg.cycle_time),
+                        ],
+                        dtype=np.float32,
+                    ),
                     "3D_command": np.array([x_vel_cmd, y_vel_cmd, yaw_vel_cmd], dtype=np.float32),
                     "joint_pos": cur_pos_obs.astype(np.float32),
                     "joint_vel": cur_vel_obs.astype(np.float32),
-                    "actions": actions.astype(np.float32),
+                    "prev_actions": actions.astype(np.float32),
                     "ang_vel": omega.astype(np.float32),
                     "euler_rotation": eu_ang.astype(np.float32),
                     "buffer": hist_obs.astype(np.float32)
                 })
 
+            positions, actions, hist_obs = policy.run(None, input_data)
+            target_q = positions
+
         # Generate PD control
         tau = pd_control(target_q, q, kps, dq, kds, default)  # Calc torques
         tau = np.clip(tau, -tau_limit, tau_limit)  # Clamp torques
@@ -326,6 +331,7 @@ def parse_modelmeta(
             dt=0.001,
             decimation=10,
             tau_factor=4.0,
+            cycle_time=policy_cfg.cycle_time,
         )
     elif args.embodiment == "stompymicro":
         policy_cfg.cycle_time = 0.2
@@ -334,16 +340,27 @@ def parse_modelmeta(
             dt=0.001,
             decimation=10,
             tau_factor=2,
+            cycle_time=policy_cfg.cycle_time,
         )
 
     if args.load_model.endswith(".onnx"):
         policy = ort.InferenceSession(args.load_model)
     else:
-        policy = convert_model_to_onnx(args.load_model, policy_cfg, save_path="policy.onnx")
+        policy = convert_model_to_onnx(
+            args.load_model, policy_cfg, save_path="policy.onnx"
+        )
 
     model_info = parse_modelmeta(
         policy.get_modelmeta().custom_metadata_map.items(),
         verbose=True,
     )
 
-    run_mujoco(args.embodiment, policy, cfg, model_info, args.keyboard_use, args.log_h5, args.render)
+    run_mujoco(
+        args.embodiment,
+        policy,
+        cfg,
+        model_info,
+        args.keyboard_use,
+        args.log_h5,
+        args.render,
+    )