Multi-Agent Reinforcement Learning for Compute Allocation in Processing Networks

Processing networks are multi-agent systems of smart sensors — each with sensing, compute, and communication capabilities — that refine data locally before sending it to a workstation for global monitoring and decision-making, common in IoT and edge-computing settings such as collaborative manufacturing robots. Each sensor faces a latency/accuracy trade-off in how much it processes onboard before transmitting.

This thesis uses multi-agent reinforcement learning to find an efficient policy for how much each sensor should refine its data before transmission, focusing on global estimation and monitoring with use cases from robotic mapping and coverage.