Working Details

Phase 1: Physics Kernel & Muscle Voxels: Implemented a virtual environment where each robot is modelled as a mass‑spring lattice. Frequency‑controlled actuation of linked spring–mass blocks formed muscle‑like voxel groups whose material constants were parameterised for mutation. Force calculation, collision detection, and time‑step integration were coded in native C++ to maximise throughput.
Phase 2: Evolutionary Strategy: Applied crossover, mutation, and fitness‑proportional selection to thousands of genomes. Fitness was defined as forward displacement per unit time; the algorithm progressively linearised motion using SVAJ‑derived functions to minimise energy bleed. Tackled the travelling‑salesman benchmark to calibrate crossover speed bumps and confirm evolutionary convergence behaviour.
Phase 3: Morphology Optimisation: Enabled symmetrical tetrahedral patterns and local material differentiation. Nearby epicentres evolved stiffness and damping properties that promoted longer stride length, producing robots that self‑organised efficient gaits after tens of thousands of generations.
Phase 4: GUI & Analytics: Combined a Java OpenGL viewer with Python dashboards for real‑time fitness tracking. Robots’ centre‑of‑mass pathways, spring tensions, and energy metrics were logged for post‑run visualisation, informing iterative genome‑operator tuning.