Do robots that learn at a door drift to going first when collisions are cheap?
research/society-learnrobots
Do robots that learn at a door drift to going first when collisions are cheap? No: with 30% robots they settle on a mix whose go share falls smoothly with collision cost (0.51 at k = 1.5, 0.03 at k = 2.8), whatever they start from, at a higher crowd cost than a fixed yielding fleet for k of 1 or more.
What it found
Robots in the go-or-yield door game of society-doorway learn their go probability from their own payoffs (action-value learning, logit policy) instead of following a fixed rule. With 30% robots, w = d = 1, they do not collapse to always going below the society-doorcost threshold k* = 2.80: the fleet go share falls smoothly with the collision cost (0.51 at k = 1.5, 0.18 at k = 2, 0.03 at k = 2.8 for response sharpness 16), regardless of whether it starts yielding or going. The learned fleet costs the crowd more than a fixed always-yield fleet for k ≥ 1. A logit quantal-response fixed point matches the simulation for blunt learners (within 0.09) but overstates going for sharp ones (0.74 predicted against 0.51 simulated at k = 1.5); the cause is untested. Learner step size and sharpness are design parameters, swept not measured; costs are free parameters.
A figure from its own output
Paper, code and results
More in Robot societies among people
- How many line-cutting robots does it take to break a queueing norm?How many line-cutting robots break a queueing norm? Mean-field critical share 12.7% (memory 9, lone-cut objection 0.9), simulation agrees; depends mostly on how fast objection fades. Stylised: objection strength is swept, not sourced.
- Does the observed rate of objection to line-cutting keep a queueing norm in a robot crowd?Do real objection rates keep a queue norm? People alone need to stop a lone cut at least 0.786 of the time; Milgram's field reactions (about 0.5, secondary source) imply 7.5% enforcer robots even with no cutters. Inputs not read from the paper; p0 is swept.
- Does a signal-reading door fleet help people once robots are not counted, and for whom?Does a signal-reading door fleet help people once robots are not counted? Only if the urgent cue costs them about a third of what it costs others, or a planner weights urgent waiting; an equal-cost cue hurts people at every weight tried.
- Can a local mean-field predict which robot blocks tip a ring crowd's passing-side convention?A deterministic local mean-field agrees with the society-allies simulation on tip/no-tip in 145 of 160 cells and is almost never too optimistic, but it overstates the share of evenly spread robots needed (0.20 against 0.12 at k = 2).
- Is the mean-field's overshoot on the ring crowd caused by finite-size noise or by memory correlation?Neither alone: a noisy memoryless map tips at 0.06 robot share for spread robots, the simulation at 0.10 and the deterministic mean-field at 0.18 (k = 2).
- Does a measured walking-side preference pin down how many robots tip a crowd's passing side?Fitting the stubborn-bloc model to the side preference Moussaïd et al. (2009) measured for two head-on walkers (about 17.5% left, read from their Fig. 4) puts the mean-field tipping robot share at 0.075 to 0.096 for memory 9 across error rates 0.02 to 0.30, against 0.217 for an unbiased crowd; memory matters more (0.033 to 0.053 at 5, 0.133 to 0.149 at 17) and a stubborn-right bloc raises it.
All 31 Robot societies among people 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.