from dotenv import load_dotenv assert load_dotenv(), "Failed to load .env file. Make sure it exists and is properly formatted." import dataclasses import enum import logging import pathlib import time import os import numpy as np from openpi_client import websocket_client_policy as _websocket_client_policy import polars as pl import rich import tqdm import tyro logger = logging.getLogger(__name__) class EnvMode(enum.Enum): """Supported environments.""" ALOHA = "aloha" ALOHA_SIM = "aloha_sim" DROID = "droid" LIBERO = "libero" HFM = "hfm" @dataclasses.dataclass class Checkpoint: """Load a policy from a trained checkpoint.""" # Training config name (e.g., "pi0_aloha_sim"). config: str # Checkpoint directory (e.g., "checkpoints/pi0_aloha_sim/exp/10000"). dir: str @dataclasses.dataclass class Args: """Command line arguments.""" # Host and port to connect to the server. host: str = "0.0.0.0" # Port to connect to the server. If None, the server will use the default port. port: int | None = 8000 # API key to use for the server. api_key: str | None = None # # Number of steps to run the policy for. # num_steps: int = 20 # Path to save the timings to a parquet file. (e.g., timing.parquet) timing_file: pathlib.Path | None = None # Environment to run the policy in. env: EnvMode = EnvMode.HFM task: str = "Pick_toys_into_box_and_lift_and_turn_and_put_on_the_chair_new_target_yaw" # policy: Checkpoint = dataclasses.field(default_factory=Checkpoint) class TimingRecorder: """Records timing measurements for different keys.""" def __init__(self) -> None: self._timings: dict[str, list[float]] = {} def record(self, key: str, time_ms: float) -> None: """Record a timing measurement for the given key.""" if key not in self._timings: self._timings[key] = [] self._timings[key].append(time_ms) def get_stats(self, key: str) -> dict[str, float]: """Get statistics for the given key.""" times = self._timings[key] return { "mean": float(np.mean(times)), "std": float(np.std(times)), "p25": float(np.quantile(times, 0.25)), "p50": float(np.quantile(times, 0.50)), "p75": float(np.quantile(times, 0.75)), "p90": float(np.quantile(times, 0.90)), "p95": float(np.quantile(times, 0.95)), "p99": float(np.quantile(times, 0.99)), } def print_all_stats(self) -> None: """Print statistics for all keys in a concise format.""" table = rich.table.Table( title="[bold blue]Timing Statistics[/bold blue]", show_header=True, header_style="bold white", border_style="blue", title_justify="center", ) # Add metric column with custom styling table.add_column("Metric", style="cyan", justify="left", no_wrap=True) # Add statistical columns with consistent styling stat_columns = [ ("Mean", "yellow", "mean"), ("Std", "yellow", "std"), ("P25", "magenta", "p25"), ("P50", "magenta", "p50"), ("P75", "magenta", "p75"), ("P90", "magenta", "p90"), ("P95", "magenta", "p95"), ("P99", "magenta", "p99"), ] for name, style, _ in stat_columns: table.add_column(name, justify="right", style=style, no_wrap=True) # Add rows for each metric with formatted values for key in sorted(self._timings.keys()): stats = self.get_stats(key) values = [f"{stats[key]:.1f}" for _, _, key in stat_columns] table.add_row(key, *values) # Print with custom console settings console = rich.console.Console(width=None, highlight=True) console.print(table) def write_parquet(self, path: pathlib.Path) -> None: """Save the timings to a parquet file.""" logger.info(f"Writing timings to {path}") frame = pl.DataFrame(self._timings) path.parent.mkdir(parents=True, exist_ok=True) frame.write_parquet(path) import torch from PIL import Image def pt_to_pil(x, normalizee=False): s, b = (0.5, 0.5) if normalizee else (1.0, 0.0) return Image.fromarray( (((x.float() * s + b).clamp(0, 1))*255.0).permute(1,2,0).cpu().numpy().astype(np.uint8) ) def obs_fn(sample, prompt): return { "observation/image": pt_to_pil(sample["observation.images.egocentric"]), # "observation/arm_joints": sample["observation.arm_joints"].numpy(), # "observation/hand_joints": sample["observation.hand_joints"].numpy(), # "observation/leg_joints": sample["observation.leg_joints"].numpy(), # "observation/torso_rpy": sample["observation.prev_rpy"].numpy(), # "observation/base_height": sample["observation.prev_height"].numpy(), "states": sample["states"].numpy()[:28], "prompt": f"g1/{prompt}" #"g1/Remove_the_cap_turn_on_the_faucet_and_fill_the_bottle_with_water", } def main(args: Args) -> None: from lerobot.common.datasets.lerobot_dataset import HF_LEROBOT_HOME from lerobot.common.datasets.lerobot_dataset import LeRobotDataset import lerobot.common.datasets.lerobot_dataset as lerobot_dataset # ds = LeRobotDataset("pick_n_squat", root="/hfm/pick_n_squat") root_dir = f"{os.environ['DATA_HOME']}/simple/G1WholebodyBendPick-v0-psi0" action_horizon = 30 action_sequence_keys = ("action",) meta = lerobot_dataset.LeRobotDatasetMetadata(root_dir) delta_timestamps={ key: [t / meta.fps for t in range(action_horizon)] for key in action_sequence_keys } dataset = LeRobotDataset(root_dir, delta_timestamps=delta_timestamps) episode_idx = 0 from_idx:int = dataset.episode_data_index["from"][episode_idx].item() to_idx:int = dataset.episode_data_index["to"][episode_idx].item() import dataclasses import enum import logging import socket import numpy as np import tyro import numpy as np import einops from openpi.policies import policy as _policy from openpi.policies import policy_config as _policy_config from openpi.serving import websocket_policy_server from openpi.training import config as _config # @dataclasses.dataclass # class Args2: # """Arguments for the serve_policy script.""" # # Environment to serve the policy for. This is only used when serving default policies. # env: EnvMode = EnvMode.HFM # # If provided, will be used in case the "prompt" key is not present in the data, or if the model doesn't have a default # # prompt. # default_prompt: str | None = "g1/Remove_the_cap_turn_on_the_faucet_and_fill_the_bottle_with_water" # # Port to serve the policy on. # port: int = 8000 # # Record the policy's behavior for debugging. # record: bool = False # # Specifies how to load the policy. If not provided, the default policy for the environment will be used. # policy: Checkpoint = dataclasses.field(default_factory=Checkpoint) # args2 = Args2( # policy = Checkpoint( # config="pi05_finetune_pick_n_squat", # dir="/hfm/cache/openpi/checkpoints/pi05_finetune_pick_n_squat/pick_n_squat/20000" # ) # ) # policy = _policy_config.create_trained_policy( # _config.get_config(args2.policy.config), args2.policy.dir, default_prompt=args2.default_prompt # ) policy = _websocket_client_policy.WebsocketClientPolicy( host=args.host, port=args.port, api_key=args.api_key, ) logger.info(f"Server metadata: {policy.get_server_metadata()}") timing_recorder = TimingRecorder() l1_losses = [] for i in tqdm.tqdm(range(from_idx, to_idx + 1, 10)): sample = dataset[i] # print(sample.keys()) # print(type(sample["action"])) # torch.Tensor, action inference_start = time.time() # print(f"============={i}=============") obs = obs_fn(sample, prompt=args.task) # print(obs["observation/image"].save(f"debug_obs_{i}.png")) obs["observation/image"] = np.array(obs["observation/image"], dtype=np.uint8) result = policy.infer(obs) pred_actions = result["actions"] # print(f"actions.shape {pred_actions.shape}") n_action_dim = pred_actions.shape[-1] gt_action = sample["action"][:, :n_action_dim].numpy() # FIXME # print(f"gt_action.shape:{gt_action.shape}") l1_loss = np.abs(gt_action - pred_actions) timing_recorder.record("client_infer_ms", 1000 * (time.time() - inference_start)) # print(l1_loss) # print("="*20) l1_losses.append(l1_loss) print("==== Final L1 Loss ====") l1_losses = np.array(l1_losses) print(f"L1 Loss shape: {l1_losses.shape}") np.save("hfm_pick_n_squat_l1_loss.npy", l1_losses) # print(f"Mean L1 Loss: {np.mean(l1_losses)}") # print(f"Std L1 Loss: {np.std(l1_losses)}" plot_error(l1_losses) def plot_error(l1_losses=None): import numpy as np import matplotlib.pyplot as plt if l1_losses is None: l1_losses = np.load("hfm_pick_n_squat_l1_loss copy.npy") print(f"L1 Loss shape: {l1_losses.shape}") print(f"Mean L1 Loss: {np.mean(l1_losses)}") if True: errors_roll = np.rad2deg(l1_losses[:, 0, 28]) errors_pitch = np.rad2deg(l1_losses[:, 0, 29]) errors_yaw = np.rad2deg(l1_losses[:, 0, 30]) # Plot plt.figure(figsize=(8, 4)) # plt.plot(np.rad2deg(errors_roll), marker='o') plt.plot(errors_roll, label="Roll (0,0)") plt.plot(errors_pitch, label="Pitch (0,1)") plt.plot(errors_yaw, label="Yaw (0,2)") plt.xlabel("Index") plt.ylabel("L1 Loss") plt.title("L1 Loss for Roll, Pitch, Yaw (l1_losses[:, 0, *]) (Degree) ") plt.grid(True) plt.legend() # plt.show() plt.savefig("l1_loss_plot_rpy.png", dpi=300, bbox_inches='tight') plt.close() if True: errors_height = l1_losses[:, 0, 31] # Plot plt.figure(figsize=(8, 4)) plt.plot(errors_height, marker='o') plt.xlabel("Index") plt.ylabel("L1 Loss") plt.title("L1 Loss for height (Meter)") plt.grid(True) plt.legend() # plt.show() plt.savefig("l1_loss_plot_height.png", dpi=300, bbox_inches='tight') plt.close() if True: errors_vx = l1_losses[:, 0, 32] # Plot plt.figure(figsize=(8, 4)) plt.plot(errors_vx, marker='o') plt.xlabel("Index") plt.ylabel("L1 Loss") plt.title("L1 Loss for Vx (m/s)") plt.grid(True) plt.legend() # plt.show() plt.savefig("l1_loss_plot_vx.png", dpi=300, bbox_inches='tight') plt.close() if True: errors_vy = l1_losses[:, 0, 33] # Plot plt.figure(figsize=(8, 4)) plt.plot(errors_vy, marker='o') plt.xlabel("Index") plt.ylabel("L1 Loss") plt.title("L1 Loss for Vy (m/s)") plt.grid(True) plt.legend() # plt.show() plt.savefig("l1_loss_plot_vy.png", dpi=300, bbox_inches='tight') plt.close() if True: errors_vyaw = l1_losses[:, 0, 34] # Plot plt.figure(figsize=(8, 4)) plt.plot(errors_vyaw, marker='o') plt.xlabel("Index") plt.ylabel("L1 Loss") plt.title("L1 Loss for Vyaw (m/s)") plt.grid(True) plt.legend() # plt.show() plt.savefig("l1_loss_plot_vyaw.png", dpi=300, bbox_inches='tight') plt.close() if True: errors_target_yaw = np.rad2deg(l1_losses[:, 0, 35]) # Plot plt.figure(figsize=(8, 4)) plt.plot(errors_target_yaw, marker='o') plt.xlabel("Index") plt.ylabel("L1 Loss") plt.title("L1 Loss for Target Yaw (Degree)") plt.grid(True) plt.legend() # plt.show() plt.savefig("l1_loss_plot_target_yaw.png", dpi=300, bbox_inches='tight') plt.close() if True: errors_arm = np.rad2deg(l1_losses[:, 0, 14:28]) plt.figure(figsize=(8, 4)) # Plot all 14 curves plt.figure(figsize=(10, 5)) for i in range(errors_arm.shape[1]): plt.plot(errors_arm[:, i], label=f"Arm Joint {i}") plt.xlabel("Index") plt.ylabel("L1 Loss") plt.title("Arm Joint L1 Loss (joints 0–13) (Degree)") plt.grid(True) plt.legend(ncol=2, fontsize=8) # plt.show() plt.savefig("l1_loss_plot_arm.png", dpi=300, bbox_inches='tight') plt.close() if True: errors_hand = np.rad2deg(l1_losses[:, 0, 0:14]) plt.figure(figsize=(8, 4)) # Plot all 14 curves plt.figure(figsize=(10, 5)) for i in range(errors_hand.shape[1]): plt.plot(errors_hand[:, i], label=f"Hand Joint {i}") plt.xlabel("Index") plt.ylabel("L1 Loss") plt.title("Hand Joint L1 Loss (joints 0–13) (Degree)") plt.grid(True) plt.legend(ncol=2, fontsize=8) # plt.show() plt.savefig("l1_loss_plot_hand.png", dpi=300, bbox_inches='tight') plt.close() if __name__ == "__main__": logging.basicConfig(level=logging.INFO) main(tyro.cli(Args)) # plot_error()