Is MuJoCo's refsafe clamp what breaks a moving grasp at a coarse timestep?

What it found

A follow-up to robofab-solreftimestep, which only read the documentation on the refsafe flag. The same jaw-force test (cover part, 400 and 800 m/s²) is run over contact time constants of 0.5 to 0.1 ms and timesteps of 0.4 to 0.05 ms with the flag on and off. 11 of 32 cells hold with it on and 10 of 32 with it off, and the two settings disagree in 7 cells. With it off, 10 runs end in a solver instability, so the clamp protects some cells and costs others, and the non-monotone timestep behaviour survives with it off. Stylised: one part, upward acceleration only, rigid box pads, no arm.

Paper, code and results

• README and method
• White paper
• Experiments and results
• Code
• Tests

More in Robot manufacturing loops

• Does an off-centre centre of mass need a harder grasp?Does an off-centre centre of mass need a harder grasp? Ball-pad elliptic-cone bound matches MuJoCo within 1.1% (5× force at 20 mm offset); flat-pad MuJoCo lies between an equal-split and a best-split corner-contact bound (stylised).
• Does the friction-grasp bound survive lift acceleration and an off-centre centre of mass?Does the friction-grasp bound survive lift acceleration and an off-centre centre of mass? Scaled bound matches MuJoCo within 6% over 66 cases; the torque bracket for an off-centre bar is inconclusive and depends on the rotation tolerance (stylised).
• Is the off-centre bar result in robofab-liftload a grasp effect or solver creep?Is the off-centre bar result in robofab-liftload a grasp effect or solver creep? The bar creeps at about 3.7 deg/s from 0.3 to 3 N in MuJoCo, with a rate set by contact settings, so the earlier verdict reflects the simulator (stylised).
• Does a rate-based "holds" criterion make a MuJoCo grasp verdict independent of solver settings?Does a rate-based "holds" criterion make a MuJoCo grasp verdict independent of solver settings? No: the minimum holding force still varies 33x to 400x or more across contact settings, like a 1 degree tolerance (negative, stylised).
• Is any rate threshold settings-robust for a MuJoCo grasp "holds" verdict?Is any rate threshold settings-robust for a MuJoCo "holds" verdict? Spread falls with the allowed tip speed and is 1.6 only at a loose 20 mm/s (0.5 s, 10 mm); 242.7 to 400 at 2 mm/s or less (stylised, preliminary).
• Does the loose 20 mm/s “holds” threshold stay settings-robust at other friction and under lift?Does the loose 20 mm/s "holds" threshold stay settings-robust at other friction and under lift? It is the only threshold under 5x spread in all six conditions (1.6 to 4.5), every other is at least 20x (stylised, preliminary).

All 53 Robot manufacturing loops projects →

Other research topics

Proper scoring & elicitation28Effort & peer prediction9Verification games & slashing19Detection & tolerances14Markets as routers13Prediction-market design9Paying contributors8Decentralised training systems23Digital twins & sim-to-real164LLM chip design loops40Robot manufacturing loops53Ambient AI assistant46Self-replicating manufacturing34Simulation-first design64Engineered clothing45Task-specific AI accelerators40Algorithmic chip-design loops42Open-source purpose-built robots44The software-engineering loop43Adversarial human imitation41Open-source residential construction41Autonomous multi-angle capture34Sensing, imaging and 3D31Energy supply and dense energy storage39Robot societies among people31Agentic co-design of low-cost robots27Laundry-hanging robots25Machine learning on qualitative data23Robots on the building site27Civilizations & group psychology1History & human sciences1

Produced with AI assistance (Claude). Results are from the code in this repository and are simulated or stylised unless the abstract says otherwise; nothing here implies affiliation with or endorsement by any lab, researcher or company cited. MIT licensed.