Build · actuator lab
Swap an actuator, see what breaks
Every number below comes from a datasheet or from Dropbear's motor-twin simulation. Pick a motor for any joint group and the robot replays the real stand, walk and wave rollouts against it: torque, speed, heat, fit in millimetres, and what it costs.
Twin: Isaac Lab 2.2 closed-loop motor twin (hw_motor_specs.py), 22 motors, 200 Hz explicit PD. Simulation only; nothing here has run on hardware. Baseline torques were produced by policies trained on the hw_v1 motors. Prices as listed on 2026-10-10.
Dots: Stand · Walk · Wave (arm raise) · Rough-terrain walk. Gate = dropbear-wbc hardware gate: RMS ≥ 1× rated warns, ≥ 1.3× fails; 5% / 20% of samples outside the envelope; p99 speed ≥ 80% / 95% of no-load.
Published deployable walk: Kimodo G1 walk, ~1.3 m/s, 0 falls, hardware gate PASS. Click an actuator in 3D to inspect it.
Knee · 2 motors · now EPS-CEM-60 cycloid (provisional)
Torque–speed · Walk
━ peak envelope ━ rated ┅ current motor · dots = every simulated sample, red = outside the envelope
Mechanical fit · current → selected
| envelope Ø | — | ||
| axial length | — | ||
| mass each | 1.2 kg | ||
| price each | — |
- output flange
- —
- rear mount
- —
Provisional twin model: no datasheet exists for this motor.
Best fit for knee — every actuator, every scenario
sorted: passes all scenarios → doesn't break other joints (mass) → cheapest → lightest
Running the simulations…
How these numbers are made
Requirements are the per-motor torque the controller demanded, and the speed it ran at, in Hyperspawn/dropbear-wbc's closed-loop motor twin (Isaac Lab 2.2 closed-loop motor twin (hw_motor_specs.py), 22 motors, 200 Hz explicit PD). They are in motor space, so the knee four-bar, the parallel ankle and the elbow linkage stay exactly as built when you swap a motor. Stand, walk and wave replay the recorded rollouts; rough terrain has no recorded poses, so only its numbers are shown.
Swapping a motor changes the load three ways, each corrected: reflected rotor inertia (rotor inertia × ratio² × the simulated acceleration), friction (datasheet back-drive torque, using the twin's friction law), and added mass (1 kg placed at each motor's housing on the serial model, weight carried through the stance feet, mapped to every motor through the measured linkage Jacobians). Added mass is applied conservatively (it always adds load); mass savings are shown but not credited.
Capability is each actuator's output torque–speed envelope in the twin's own form: peak torque capped by the voltage line. For the three motors on the robot today the twin's validated parameters are used. For catalog motors the line comes from the vendor's 48 V test curve where the sheet has one, otherwise from the rated point (conservative). Rebuilt from the transcribed datasheets, this reproduces the twin's X10 parameters exactly.
Limits. Simulation only; nothing here has run on hardware. Baseline torques were produced by policies trained on the hw_v1 motors. A large change of motor (especially speed or reflected inertia) means re-training and re-validating the policy in dropbear-wbc before trusting it on hardware. Dimensions are from each manufacturer's installation drawing; the 3D envelope is placed coaxially on the joint, its axial position is approximate.