File size: 5,323 Bytes
c4f8e4a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 | from __future__ import annotations
import argparse
import json
import subprocess
import sys
import time
from pathlib import Path
import numpy as np
import pandas as pd
def parse_args() -> argparse.Namespace:
parser = argparse.ArgumentParser()
parser.add_argument("--run-dir", required=True)
parser.add_argument("--ckpt-step", required=True)
parser.add_argument("--device", default="cuda:0")
parser.add_argument("--prompt", required=True)
parser.add_argument("--motion-name", default="psi0_yth_v4_test")
parser.add_argument("--num-output-frames", type=int, default=300)
parser.add_argument("--num-inference-steps", type=int, default=8)
parser.add_argument("--start-index", type=int, default=0)
parser.add_argument("--sonic-repo", default="/cpfs01/thscc/sharestorage/AIGC-06-01/Hithink_VideoGen_Group/user_workspace/yetinghong/code/GR00T-WholeBodyControl")
parser.add_argument("--output-root", required=True)
parser.add_argument("--run-sonic", default="1")
return parser.parse_args()
def run_cmd(cmd: list[str], cwd: Path) -> None:
print("\n[run]")
print(" ".join(cmd))
subprocess.run(cmd, cwd=str(cwd), check=True)
def read_csv_array(path: Path) -> np.ndarray:
if not path.exists():
raise FileNotFoundError(path)
return pd.read_csv(path).to_numpy(dtype=np.float64)
def validate_motion_dir(motion_dir: Path) -> None:
joint_pos = read_csv_array(motion_dir / "joint_pos.csv")
joint_vel = read_csv_array(motion_dir / "joint_vel.csv")
body_pos = read_csv_array(motion_dir / "body_pos.csv")
body_quat = read_csv_array(motion_dir / "body_quat.csv")
body_lin_vel = read_csv_array(motion_dir / "body_lin_vel.csv")
body_ang_vel = read_csv_array(motion_dir / "body_ang_vel.csv")
frames = joint_pos.shape[0]
checks = [
("joint_pos", joint_pos, 29),
("joint_vel", joint_vel, 29),
("body_pos", body_pos, 3),
("body_quat", body_quat, 4),
("body_lin_vel", body_lin_vel, 3),
("body_ang_vel", body_ang_vel, 3),
]
for name, array, width in checks:
if array.shape[0] != frames or array.shape[1] != width:
raise ValueError(f"{name} shape {array.shape} does not match expected ({frames}, {width})")
quat_norm = np.linalg.norm(body_quat, axis=1)
if not np.all(np.isfinite(joint_pos)):
raise ValueError("joint_pos contains non-finite values")
if not np.all(np.isfinite(body_quat)):
raise ValueError("body_quat contains non-finite values")
if np.max(np.abs(quat_norm - 1.0)) > 1e-2:
raise ValueError("body_quat is not normalized enough")
print(f"[check] SONIC reference motion OK: {motion_dir}")
print(f"[check] frames={frames}, fps=50, duration={frames / 50.0:.2f}s")
def main() -> None:
args = parse_args()
psi_repo = Path(__file__).resolve().parents[2]
sonic_repo = Path(args.sonic_repo).resolve()
run_tag = time.strftime("%y%m%d%H%M%S")
output_root = Path(args.output_root).resolve() / run_tag
motion_dir = output_root / "g1" / "csv_sonic" / "manual" / args.motion_name
# <modify> 先复用仓库里已有的TextOp->36D->SONIC CSV导出逻辑,避免重新写一套动作解码。
infer_cmd = [
sys.executable,
"scripts/infer/seed_text_to_motion.py",
"--run-dir",
args.run_dir,
"--ckpt-step",
args.ckpt_step,
"--output-dir",
str(output_root),
"--prompt",
args.prompt,
"--prompt-motion-name",
args.motion_name,
"--num-output-frames",
str(args.num_output_frames),
"--num-inference-steps",
str(args.num_inference_steps),
"--start-index",
str(args.start_index),
"--device",
args.device,
]
run_cmd(infer_cmd, psi_repo)
validate_motion_dir(motion_dir)
meta = {
"prompt": args.prompt,
"run_dir": args.run_dir,
"ckpt_step": args.ckpt_step,
"motion_dir": str(motion_dir),
"sonic_repo": str(sonic_repo),
"target_fps": 50,
}
(output_root / "psi0_sonic_bridge_meta.json").write_text(
json.dumps(meta, ensure_ascii=False, indent=2),
encoding="utf-8",
)
if args.run_sonic != "1":
print("[done] Generated motion only. Set RUN_SONIC=\"1\" in sh to run tracking.")
return
sonic_python = sonic_repo / ".venv_sim" / "bin" / "python"
if not sonic_python.exists():
raise FileNotFoundError(f"SONIC python not found: {sonic_python}")
sonic_output = sonic_repo / "outputs" / "psi0_yth_v4_motion_tracking" / run_tag
video_name = args.motion_name + ".mp4"
# <modify> 这里调用GR00T-WholeBodyControl现有headless reference motion runner,实际进入SONIC motion tracking。
sonic_cmd = [
str(sonic_python),
"gear_sonic/scripts/run_headless_reference_motion.py",
"--motion-path",
str(motion_dir),
"--output-dir",
str(sonic_output),
"--video-name",
video_name,
"--video-width",
"1440",
"--video-height",
"832",
"--video-fps",
"15",
]
run_cmd(sonic_cmd, sonic_repo)
print(f"[done] SONIC output: {sonic_output}")
if __name__ == "__main__":
main()
|