File size: 3,292 Bytes
6aaad38 | 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 | """Is the arm slow, or does it dither? Closed-loop right-arm kinematics per execution mode.
Reads episodes logged with chunk_eval.py --log-trajectory (rows: step, 6 right-arm joint angles,
gripper-cube distance) and summarizes the approach phase (steps 0-150):
speed joint-space path length per step (rad/step)
efficiency net joint displacement / path length (1 = straight path, low = back and forth)
reversals share of steps where a joint's velocity changes sign (joints moving > 1e-4 rad/step)
dist@100/150 gripper-cube distance (m) at steps 100 and 150
Usage:
python scripts/analyze_trajectories.py outputs/hub/results/diag_official outputs/hub/results/diag_ours100k \
--out results/analysis/trajectories.json
"""
import argparse
import glob
import json
from pathlib import Path
import numpy as np
T_END = 150
ORDER = ["full_d00_m0", "full_d10_m0", "replan25_d00_m0", "replan10_d00_m0", "replan10_d10_m0"]
def episode_metrics(traj: list[list[float]]) -> dict | None:
rows = np.array([r for r in traj if r[0] <= T_END])
if len(rows) < T_END: # ended early (success before step 150) or not logged
return None
q, dist = rows[:, 1:7], rows[:, 7]
dq = np.diff(q, axis=0)
step = np.linalg.norm(dq, axis=1)
path = step.sum()
moving = np.abs(dq[1:]) > 1e-4
flips = (np.sign(dq[1:]) != np.sign(dq[:-1])) & moving & (np.abs(dq[:-1]) > 1e-4)
return {
"speed": float(step.mean()),
"efficiency": float(np.linalg.norm(q[-1] - q[0]) / path) if path > 0 else float("nan"),
"reversals": float(flips.sum() / max(moving.sum(), 1)),
"dist100": float(dist[100]),
"dist150": float(dist[T_END]),
}
def main() -> None:
parser = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
parser.add_argument("runs", nargs="+", type=Path)
parser.add_argument("--out", type=Path, required=True)
args = parser.parse_args()
summary = {}
for run in args.runs:
summary[run.name] = {}
print(f"== {run.name} (median over episodes; n = episodes still running at step {T_END})")
print(f" {'condition':17s} {'n':>3s} {'speed':>7s} {'effic.':>7s} {'revers.':>8s} {'d@100':>6s} {'d@150':>6s} {'success':>8s}")
for cond in ORDER:
eps = [e for f in glob.glob(str(run / cond / "seeds_*.json")) for e in json.load(open(f))["episodes"]]
if not eps:
continue
ms = [m for e in eps if (m := episode_metrics(e.get("trajectory", [])))]
med = {k: float(np.median([m[k] for m in ms])) for k in ms[0]} if ms else {}
med["n"] = len(ms)
med["success_rate"] = sum(e["success"] for e in eps) / len(eps)
summary[run.name][cond] = med
print(f" {cond:17s} {med['n']:3d} {med.get('speed', float('nan')):7.4f} {med.get('efficiency', float('nan')):7.2f} "
f"{med.get('reversals', float('nan')):8.3f} {med.get('dist100', float('nan')):6.3f} {med.get('dist150', float('nan')):6.3f} "
f"{100 * med['success_rate']:7.0f}%")
args.out.parent.mkdir(parents=True, exist_ok=True)
args.out.write_text(json.dumps(summary, indent=1))
if __name__ == "__main__":
main()
|