FAIL
All revised hip axes clear, but unchanged knee and ankle axes still exceed the saturation threshold. The run is retained as a rejected controller-and-model configuration.
This is a bounded ideal-fixture result—not physical walking validation.
WS-L01
53.70%
Maximum rotary saturation; baseline 84.7%. Tracking error 1.09°.
FAIL SUCCESSOR IDEAL-FIXTURE NUMERICAL SCREENWS-R01
51.59%
Maximum rotary saturation; baseline 84.6%. Tracking error 1.09°.
FAIL SUCCESSOR IDEAL-FIXTURE NUMERICAL SCREENPreregistered transmission scenario
| Axis | Baseline | Successor | Current endpoint | Inverse peak | No-load speed reserve |
|---|---|---|---|---|---|
| L_HIP_PITCH | 1.500:1 | 4.000:1 | 17.17 N·m | 11.45 N·m | 3.13× |
| L_HIP_ROLL | 2.000:1 | 4.000:1 | 17.17 N·m | 12.74 N·m | 14.98× |
| R_HIP_PITCH | 1.500:1 | 4.000:1 | 17.17 N·m | 11.46 N·m | 3.13× |
| R_HIP_ROLL | 2.000:1 | 4.000:1 | 17.17 N·m | 12.75 N·m | 14.98× |
What remains unmodeled
The 4:1 ratio is not yet a physical transmission. This run omits added pulley/shaft/bearing mass, belt compliance, backlash, reflected rotor inertia, tooth loading, efficiency variation, actuator torque-speed behavior and heat. Model-based inverse feedforward also requires robustness tests against mass, contact, latency and state-estimation error.
Evidence downloads
sequence results · all 25 axes · 50 Hz samples · scenario definition · MuJoCo model · open holds