Event

Release of a Distributed Cluster Morphogenesis Framework

1 reports · 1 sources

Ray Verma and other authors released a modular framework that combines neural cellular automata (NCA) with existing site selection and collision-aware control for swarm robot morphogenesis. Its core change is that already-placed robots identify new attachment sites by updating and sharing local NCA states, allowing the swarm to grow incrementally at deployment without a predefined site list. The authors also propose offline sparse-retention fine-tuning (simulating partial attachment) and a false-positive loss that penalizes activation of invalid sites, to bridge NCA growth and incremental physical assembly. Simulations show that multiple lattice morphologies can be generated and scaled to hundreds of robots, and that fine-tuning yields a significant improvement in morphology accuracy; Crazyflie quadrotor experiments validated the integrated execution pipeline, but the relative observations of the robots used were reconstructed by external tracking. All the above results are as reported by the authors.