PALCAS: Federated Reinforcement Learning for Priority-Aware Autonomous Lane Changes

A team of researchers has created PALCAS, an intelligent lane change advisory system for autonomous vehicles (AVs) that is aware of priorities, utilizing multi-agent federated reinforcement learning. In contrast to traditional single-agent or centralized multi-agent methods, PALCAS assesses lane changes based on the urgency of a vehicle's destination. It features a unique priority-aware safe lane-change reward function applicable to both mandatory and discretionary situations. The parameterized deep Q-network (PDQN) algorithm is utilized to facilitate agent cooperation, managing both lateral and longitudinal movements. Simulations conducted with the SUMO traffic simulator and the Mosaic V2X communication framework demonstrate that PALCAS greatly enhances traffic efficiency.