Create a minimal 'World Model' environment for reinforcement learning in physical space. Use cheap hardware (e.g., Raspberry Pi) to train agents for simple tasks.