GPU-accelerated evolutionary framework for studying morphable soft-body robots that approach, combine, and locomote as merged entities using 400-particle spring-mass systems controlled by evolved neural networks. v4 updates (March 6, 2026): Experiment 10 (Friction Sweet Spot): 7-ratio sweep confirms No-Cost Differentiation — Diff×Fit = 171.0 (1.8× mass Sweet Spot) Experiment 11 (Double Asymmetry): Body + Environment asymmetry = destructive interference (+179 → +169) Experiment 12 (Muscle Synergy): 1D center-of-mass shift overcomes DoF Trap, achieving all-time highest fitness +210 (21% above baseline) Experiment 13 (Environmental Reversal): Plasticity test — flip-trained controllers maintain +171 (−4% vs specialist) Expanded from 8 to 12 principles Key findings (12 principles): Symmetry Locks Theorem: Symmetric bodies cannot differentiate (r=0.742). Inertial Asymmetry Principle: Only mass asymmetry breaks synchronization (Mapping Inequivalence via F=ma). Transient Differentiation: Role specialization peaks early then is partially reabsorbed for efficiency. Reward Paradox: Multi-phase fitness can mask poor locomotion via reward hacking. Symmetry Optimality: Evolution chooses symmetric masses when free to evolve morphology. The Sweet Spot: Mass ratio 3:1 maximizes differentiation per unit fitness cost. Degrees of Freedom Trap: Adding control dimensions catastrophically expands the search space. Environmental Differentiation: Asymmetric environments induce differentiation in symmetric bodies at superior fitness. No-Cost Differentiation: Friction asymmetry induces differentiation without fitness cost (Diff×Fit = 171.0, 1.8× mass Sweet Spot). Asymmetry Interference: Combining body + environment asymmetry produces destructive interference (+179 → +169). Muscle Synergy Principle: 1D center-of-mass shift overcomes DoF Trap at all-time highest fitness +210 (21% above baseline). Adaptive Robustness: Flip-trained controllers maintain +171 across changing environments (−4% vs specialist). All experiments run on a single NVIDIA RTX 5080 Laptop GPU in under 5 hours total.
Hiroto Funasaki (Fri,) studied this question.