code-GMR / scripts /bvh_to_robot.py
Tinghong-Ye22's picture
Add files using upload-large-folder tool
f6aca0c verified
Raw History Blame Contribute Delete
5.16 kB
import argparse
import pathlib
import time
from general_motion_retargeting import GeneralMotionRetargeting as GMR
from general_motion_retargeting import RobotMotionViewer
from general_motion_retargeting.utils.lafan1 import load_bvh_file
from rich import print
from tqdm import tqdm
import os
import numpy as np
if __name__ == "__main__":
HERE = pathlib.Path(__file__).parent
parser = argparse.ArgumentParser()
parser.add_argument(
"--bvh_file",
help="BVH motion file to load.",
required=True,
type=str,
)
parser.add_argument(
"--format",
choices=["lafan1", "nokov"],
default="lafan1",
)
parser.add_argument(
"--loop",
default=False,
action="store_true",
help="Loop the motion.",
)
parser.add_argument(
"--robot",
choices=["unitree_g1", "unitree_g1_with_hands", "booster_t1", "stanford_toddy", "fourier_n1", "engineai_pm01", "pal_talos"],
default="unitree_g1",
)
parser.add_argument(
"--record_video",
action="store_true",
default=False,
)
parser.add_argument(
"--video_path",
type=str,
default="videos/example.mp4",
)
parser.add_argument(
"--rate_limit",
action="store_true",
default=False,
)
parser.add_argument(
"--save_path",
default=None,
help="Path to save the robot motion.",
)
parser.add_argument(
"--motion_fps",
default=30,
type=int,
)
args = parser.parse_args()
if args.save_path is not None:
save_dir = os.path.dirname(args.save_path)
if save_dir: # Only create directory if it's not empty
os.makedirs(save_dir, exist_ok=True)
qpos_list = []
# Load SMPLX trajectory
lafan1_data_frames, actual_human_height = load_bvh_file(args.bvh_file, format=args.format)
# Initialize the retargeting system
retargeter = GMR(
src_human=f"bvh_{args.format}",
tgt_robot=args.robot,
actual_human_height=actual_human_height,
)
motion_fps = args.motion_fps
robot_motion_viewer = RobotMotionViewer(robot_type=args.robot,
motion_fps=motion_fps,
transparent_robot=0,
record_video=args.record_video,
video_path=args.video_path,
# video_width=2080,
# video_height=1170
)
# FPS measurement variables
fps_counter = 0
fps_start_time = time.time()
fps_display_interval = 2.0 # Display FPS every 2 seconds
print(f"mocap_frame_rate: {motion_fps}")
# Create tqdm progress bar for the total number of frames
pbar = tqdm(total=len(lafan1_data_frames), desc="Retargeting")
# Start the viewer
i = 0
while True:
# FPS measurement
fps_counter += 1
current_time = time.time()
if current_time - fps_start_time >= fps_display_interval:
actual_fps = fps_counter / (current_time - fps_start_time)
print(f"Actual rendering FPS: {actual_fps:.2f}")
fps_counter = 0
fps_start_time = current_time
# Update progress bar
pbar.update(1)
# Update task targets.
smplx_data = lafan1_data_frames[i]
# retarget
qpos = retargeter.retarget(smplx_data)
# visualize
robot_motion_viewer.step(
root_pos=qpos[:3],
root_rot=qpos[3:7],
dof_pos=qpos[7:],
human_motion_data=retargeter.scaled_human_data,
rate_limit=args.rate_limit,
follow_camera=True,
# human_pos_offset=np.array([0.0, 0.0, 0.0])
)
if args.loop:
i = (i + 1) % len(lafan1_data_frames)
else:
i += 1
if i >= len(lafan1_data_frames):
break
if args.save_path is not None:
qpos_list.append(qpos)
if args.save_path is not None:
import pickle
root_pos = np.array([qpos[:3] for qpos in qpos_list])
# save from wxyz to xyzw
root_rot = np.array([qpos[3:7][[1,2,3,0]] for qpos in qpos_list])
dof_pos = np.array([qpos[7:] for qpos in qpos_list])
local_body_pos = None
body_names = None
motion_data = {
"fps": motion_fps,
"root_pos": root_pos,
"root_rot": root_rot,
"dof_pos": dof_pos,
"local_body_pos": local_body_pos,
"link_body_list": body_names,
}
with open(args.save_path, "wb") as f:
pickle.dump(motion_data, f)
print(f"Saved to {args.save_path}")
# Close progress bar
pbar.close()
robot_motion_viewer.close()