Files
lerobot_piper/lerobot/scripts/eval.py

72 lines
2.2 KiB
Python

from pathlib import Path
import hydra
import imageio
import numpy as np
import torch
from termcolor import colored
from ..lib.envs import make_env
from ..lib.utils import set_seed
def eval_agent(
env, agent, num_episodes: int, save_video: bool = False, video_path: Path = None
):
"""Evaluate a trained agent and optionally save a video."""
if save_video:
assert video_path is not None
assert video_path.suffix == ".mp4"
episode_rewards = []
episode_successes = []
episode_lengths = []
for i in range(num_episodes):
obs, done, ep_reward, t = env.reset(), False, 0, 0
ep_success = False
if save_video:
frames = []
while not done:
action = agent.act(obs, t0=t == 0, eval_mode=True, step=step)
obs, reward, done, info = env.step(action.cpu().numpy())
ep_reward += reward
if "success" in info and info["success"]:
ep_success = True
if save_video:
frame = env.render(
mode="rgb_array",
# TODO(rcadene): make height, width, camera_id configurable
height=384,
width=384,
camera_id=0,
)
frames.append(frame)
t += 1
episode_rewards.append(float(ep_reward))
episode_successes.append(float(ep_success))
episode_lengths.append(t)
if save_video:
frames = np.stack(frames).transpose(0, 3, 1, 2)
video_path.parent.mkdir(parents=True, exist_ok=True)
# TODO(rcadene): make fps configurable
imageio.mimsave(video_path, frames, fps=15)
return {
"episode_reward": np.nanmean(episode_rewards),
"episode_success": np.nanmean(episode_successes),
"episode_length": np.nanmean(episode_lengths),
}
@hydra.main(version_base=None, config_name="default", config_path="../configs")
def eval(cfg: dict):
assert torch.cuda.is_available()
set_seed(cfg.seed)
print(colored("Log dir:", "yellow", attrs=["bold"]), cfg.log_dir)
env = make_env(cfg)
eval_metrics = eval_agent(env, agent, num_episodes=10, save_video=True)
if __name__ == "__main__":
eval()