tars-mini / sim /speed_study.py
tuanhqv123's picture
TARS mini: CAD, MuJoCo sim, trained policies and 5-action demo
4b7148d verified
Raw History Blame Contribute Delete
2.34 kB
"""How fast can walk_forward run on the CHEAPEST build? Domain-randomised, 24 trials per speed."""
import json, os, sys
import numpy as np
from multiprocessing import Pool
from model import Params
from sim import TarsSim
from tarsai_gaits import walk_forward
CHEAP = dict(cells=0.046, ups=0.013) # 1x18650 + holder, Seeed Lipo Rider Plus
RANGES = dict(mu=(0.5, 1.0), mass_scale=(0.9, 1.15), crank_r=(0.008, 0.016),
hip_from_top=(0.020, 0.040), p_band_deg=(3.0, 6.0))
def trial(a):
speed, stride, seed = a
rng = np.random.default_rng(seed)
kw = dict(CHEAP, **{k: float(rng.uniform(*v)) for k, v in RANGES.items()})
kw["tau_stall_kgcm"] = 1.85 * float(rng.uniform(0.85, 1.15))
s = TarsSim(Params(**kw)); s.run(0.5)
y0, n0, t0 = s.d.qpos[1], len(s.log["t"]), s.d.time
walk_forward(s, speed, stride); s.run(0.8)
fwd = -(s.d.qpos[1] - y0) * 100
tq = s.torque_stats()
sw = max(tq["sL"]["p95"], tq["sR"]["p95"])
tilt = max(s.log["up"][n0:])
ok_motion = fwd >= 3 and tilt <= 15 and s.log["up"][-1] < 5
return dict(speed=speed, stride=stride, fwd=fwd, dur=s.d.time - t0, tilt=tilt, swing_p95=sw,
I_peak=s.current_stats()["peak50ms_A"], walks=bool(ok_motion), ok=bool(ok_motion and sw <= 70))
if __name__ == "__main__":
combos = [(sp, 15) for sp in (0.5, 0.7, 0.8, 0.9, 0.95)] + [(sp, st) for sp in (0.5, 0.6) for st in (20, 25, 30)]
with Pool(9) as p:
R = p.map(trial, [(sp, st, 2000 + i) for sp, st in combos for i in range(24)])
json.dump(R, open("../results/speed_study.json", "w"), indent=1)
print("mass g:", round(Params(**CHEAP).masses()["total"] * 1000))
for sp, st in combos:
X = [r for r in R if r["speed"] == sp and r["stride"] == st]
print(f"stride {st:2d}% ", end="")
v = np.array([r["fwd"] / r["dur"] for r in X])
print(f"speed {sp:.2f}: walks {100*np.mean([r['walks'] for r in X]):5.1f}% within torque {100*np.mean([r['ok'] for r in X]):5.1f}% "
f"walk time {np.median([r['dur'] for r in X]):4.1f}s fwd {np.median([r['fwd'] for r in X]):4.1f}cm "
f"velocity {np.median(v):4.2f} cm/s max tilt {max(r['tilt'] for r in X):4.1f}° "
f"swing p95 {np.median([r['swing_p95'] for r in X]):3.0f}% I_peak {max(r['I_peak'] for r in X):.2f}A")