Hybrid Model Balances Centralized Control and Self-Organization in Morphogenesis
A research paper on arXiv proposes a hybrid model for embodied morphogenesis that combines a convolutional neural network controller with a Gray-Scott reaction-diffusion substrate. The controller modulates feed and kill parameters to guide pattern formation. The hybrid regime achieved 100% strict convergence in about 165 steps, outperforming pure RD and neural network-dominant baselines, while using approximately 15 times less L1 control effort. The study explores the division of labor between centralized guidance and distributed material dynamics.
Key facts
- Paper title: Balancing Centralized Learning and Distributed Self-Organization: A Hybrid Model for Embodied Morphogenesis
- Published on arXiv with ID 2511.10101
- Uses a compact full-resolution convolutional controller
- Controller applies smooth gain-scheduled modulations of feed and kill parameters (Delta F and Delta K)
- Hybrid regime achieved 100% strict convergence at approximately 165 steps
- Pure RD and NN-dominant baseline did not converge within the same horizon
- Hybrid regime matched substrate's spectral selectivity
- Used approximately 15 times less L1 effort
Entities
Institutions
- arXiv