gearboxAssembly / data /scripts /VLA_agent.py
introvoyz045's picture
Migrated from GitHub
8a24b46 verified
Raw History Blame Contribute Delete
4.9 kB
# Copyright (c) 2022-2025, The Isaac Lab Project Developers (https://github.com/isaac-sim/IsaacLab/blob/main/CONTRIBUTORS.md).
# All rights reserved.
#
# SPDX-License-Identifier: BSD-3-Clause
"""Script to an environment with random action agent."""
"""Launch Isaac Sim Simulator first."""
import argparse
from isaaclab.app import AppLauncher
# add argparse arguments
parser = argparse.ArgumentParser(description="Random agent for Isaac Lab environments.")
parser.add_argument(
"--disable_fabric", action="store_true", default=False, help="Disable fabric and use USD I/O operations."
)
parser.add_argument("--num_envs", type=int, default=None, help="Number of environments to simulate.")
parser.add_argument("--task", type=str, default=None, help="Name of the task.")
parser.add_argument("--checkpoint", type=str, default=None, help="Path to the VLA model checkpoint.")
# append AppLauncher cli args
AppLauncher.add_app_launcher_args(parser)
# parse the arguments
args_cli = parser.parse_args()
# launch omniverse app
app_launcher = AppLauncher(args_cli)
simulation_app = app_launcher.app
"""Rest everything follows."""
import gymnasium as gym
import torch
import isaaclab_tasks
from isaaclab_tasks.utils import parse_env_cfg
import Galaxea_Lab_External.tasks
from Galaxea_Lab_External.VLA.ACT.policy_wrapper import ACTPolicyWrapper, DiffusionPolicyWrapper, BCPolicyWrapper, DataReplayPolicyWrapper
def main():
"""Random actions agent with Isaac Lab environment."""
# create environment configuration
env_cfg = parse_env_cfg(
args_cli.task, device=args_cli.device, num_envs=args_cli.num_envs, use_fabric=not args_cli.disable_fabric
)
# Load ACT policy
# checkpoint_path = '/scratch/SM/AAAI_challenge/gearboxAssembly/source/Galaxea_Lab_External/Galaxea_Lab_External/VLA/ACT/act/ckpt/policy_best.ckpt'
if args_cli.checkpoint is None:
print("No checkpoint path provided")
exit()
else:
checkpoint_path = args_cli.checkpoint
temporal_agg = True
policy = ACTPolicyWrapper(checkpoint_path, temporal_agg=temporal_agg)
# Test policy
# qpos = torch.rand(1, 14, device=args_cli.device)
# images = torch.rand(1, 3, 3, 240, 320, device=args_cli.device)
# action = policy.predict(qpos, images)
# print(f"action: {action}")
# exit()
# create environment
env = gym.make(args_cli.task, cfg=env_cfg)
# sample_every_n_steps = max(int(sample_period / env.step_dt), 1)
print("env type: ", type(env))
# print info (this is vectorized environment)
print(f"[INFO]: Gym observation space: {env.observation_space}")
print(f"[INFO]: Gym action space: {env.action_space}")
# reset environment
obs, _ = env.reset()
# simulate environment
while simulation_app.is_running():
# run everything in inference mode
with torch.inference_mode():
# Extract observations from current state
left_arm_joint_pos = obs['policy']['left_arm_joint_pos']
right_arm_joint_pos = obs['policy']['right_arm_joint_pos']
left_gripper_joint_pos = obs['policy']['left_gripper_joint_pos']
right_gripper_joint_pos = obs['policy']['right_gripper_joint_pos']
qpos = torch.cat([left_arm_joint_pos, left_gripper_joint_pos.unsqueeze(0), right_arm_joint_pos, right_gripper_joint_pos.unsqueeze(0)], dim=-1)
head_rgb = obs['policy']['head_rgb'].unsqueeze(0).permute(0, 1, 4, 2, 3)
left_hand_rgb = obs['policy']['left_hand_rgb'].unsqueeze(0).permute(0, 1, 4, 2, 3)
right_hand_rgb = obs['policy']['right_hand_rgb'].unsqueeze(0).permute(0, 1, 4, 2, 3)
images = torch.cat([head_rgb, left_hand_rgb, right_hand_rgb], dim=1)
# Change dtype of images to float32
images = images.to(torch.float32)
# Print shape of qpos and images
print(f"qpos shape: {qpos.shape}")
print(f"images shape: {images.shape}")
# Predict action using VLA policy
action = policy.predict(qpos, images)
action_reordered = torch.cat([
action[:, :6], # L_arm
action[:, 7:13], # R_arm
action[:, 6:7], # L_grip
action[:, 13:14] # R_grip
], dim=-1)
print(f"VLA predicted action: {action_reordered}")
# Apply VLA predicted action to environment (only once per loop)
obs, reward, terminated, truncated, info = env.step(action_reordered)
print(f"Terminated: {terminated}")
print(f"Truncated: {truncated}")
if terminated or truncated:
obs, _ = env.reset()
# close the simulator
env.close()
if __name__ == "__main__":
# run the main function
main()
# close sim app
simulation_app.close()