Download code/ti5_arm_scripts/audit_velocity_feedback.py from chenhuaDr/data_calibration_robot: direct link, hf CLI and curl.
- Browser
- Download file 4.12 kB
-
https://huggingface.co/datasets/chenhuaDr/data_calibration_robot/resolve/main/code/ti5_arm_scripts/audit_velocity_feedback.py
- Command line
-
hf download hf://datasets/chenhuaDr/data_calibration_robot/code/ti5_arm_scripts/audit_velocity_feedback.py
-
curl -L -o audit_velocity_feedback.py https://huggingface.co/datasets/chenhuaDr/data_calibration_robot/resolve/main/code/ti5_arm_scripts/audit_velocity_feedback.py
4.12 kB
| #!/usr/bin/env python3 | |
| """Offline scale audit of historical HORUS captures; never connects to hardware. | |
| Replays the causal position regression on capture timestamps. Historical captures | |
| are only ~100 Hz, so this is not a validation of the 1 kHz PDO implementation. | |
| Both the replay and the reference use position, not an independent tachometer. | |
| """ | |
| import argparse | |
| from collections import deque | |
| import hashlib | |
| import json | |
| from pathlib import Path | |
| import numpy as np | |
| def audit(path): | |
| columns = ("elapsed_s", "axis", "position_rad", "velocity_rad_s", "enabled", "fault") | |
| # Recorder format: first line is a comment, second line is the CSV header. | |
| data = np.genfromtxt(path, delimiter=",", names=True, skip_header=1, | |
| usecols=columns, dtype=float) | |
| report = { | |
| "source": str(path), | |
| "sha256": hashlib.sha256(path.read_bytes()).hexdigest(), | |
| "method": "Historical capture causal 20 ms position-regression replay; " | |
| "reference is central 200 ms position difference. " | |
| "Not 1 kHz hardware validation or independent velocity calibration.", | |
| "axes": [], | |
| } | |
| for axis in sorted(np.unique(data["axis"]).astype(int)): | |
| rows = data[data["axis"] == axis] | |
| t, q = rows["elapsed_s"], rows["position_rad"] | |
| reported = rows["velocity_rad_s"] | |
| enabled = (rows["enabled"] > 0) & (rows["fault"] == 0) | |
| estimate = np.full(len(t), np.nan) | |
| history = deque() | |
| for i in range(len(t)): | |
| if i and (t[i] - t[i-1] > 0.05 or t[i] <= t[i-1]): | |
| history.clear() | |
| history.append(i) | |
| while len(history) > 2 and t[i] - t[history[1]] >= 0.02: | |
| history.popleft() | |
| if len(history) >= 3 and t[i] - t[history[0]] >= 0.002: | |
| indices = np.array(history) | |
| times = t[indices] - t[i] | |
| positions = q[indices] - q[i] | |
| times -= times.mean() | |
| positions -= positions.mean() | |
| estimate[i] = np.dot(times, positions) / np.dot(times, times) | |
| reference = (np.interp(t + 0.1, t, q) - np.interp(t - 0.1, t, q)) / 0.2 | |
| selected = (enabled & np.isfinite(estimate) & (abs(reference) > 0.03) | |
| & (t > t[0] + 0.2) & (t < t[-1] - 0.2)) | |
| # Reference windows must not cross telemetry gaps or state transitions. | |
| cuts = np.flatnonzero((np.diff(t) > 0.05) | (np.diff(t) <= 0) | |
| | (enabled[1:] != enabled[:-1])) | |
| for cut in cuts: | |
| selected[(t >= t[cut] - 0.11) & (t <= t[cut+1] + 0.11)] = False | |
| def metrics(values): | |
| if not selected.any(): | |
| return None | |
| ref = reference[selected] | |
| return { | |
| "gain_vs_position_reference": float(np.dot(ref, values[selected]) | |
| / np.dot(ref, ref)), | |
| "rmse_rad_s": float(np.sqrt(np.mean((ref - values[selected]) ** 2))), | |
| } | |
| report["axes"].append({ | |
| "axis": int(axis), "moving_samples": int(selected.sum()), | |
| "recorded_velocity": metrics(reported), | |
| "position_velocity_replay": metrics(estimate), | |
| }) | |
| return report | |
| def main(): | |
| parser = argparse.ArgumentParser(description=__doc__) | |
| parser.add_argument("--csv", type=Path, required=True) | |
| parser.add_argument("--output", type=Path, required=True) | |
| args = parser.parse_args() | |
| report = audit(args.csv) | |
| args.output.parent.mkdir(parents=True, exist_ok=True) | |
| args.output.write_text(json.dumps(report, indent=2) + "\n") | |
| for row in report["axes"]: | |
| before, after = row["recorded_velocity"], row["position_velocity_replay"] | |
| if before is not None: | |
| print(f"J{row['axis']}: recorded gain={before['gain_vs_position_reference']:.4f}, " | |
| f"position gain={after['gain_vs_position_reference']:.4f}, " | |
| f"RMSE={before['rmse_rad_s']:.5f} -> {after['rmse_rad_s']:.5f} rad/s") | |
| if __name__ == "__main__": | |
| main() | |