Insect-inspired modular architectures as inductive biases for reinforcement learning

Anne E. Staples

Most reinforcement-learning (RL) controllers used in continuous control are architecturally centralized: observations are compressed into a single latent state from which both valu…