PALCAS: A Priority-Aware Intelligent Lane Change Advisory System for Autonomous Vehicles using Federated Reinforcement Learning
New AI system uses destination urgency to decide when AVs should change lanes.
PALCAS (Priority-Aware Intelligent Lane Change Advisory System) is a new approach to autonomous vehicle lane changing, developed by researchers Yassine Ibork, Nhat Ha Nguyen, Myounggyu Won, and Lokesh Das. Unlike conventional methods that treat lane changes as single-agent or centralized multi-agent decisions, PALCAS employs federated reinforcement learning across multiple vehicles. This decentralized framework allows each AV to learn a shared policy without sharing raw data, preserving privacy while improving coordination.
The system's key innovation is a priority-aware reward function that considers each vehicle's destination urgency. This enables PALCAS to make judicious lane change decisions in both mandatory scenarios (e.g., exiting the highway) and discretionary ones (e.g., overtaking slower traffic). PALCAS uses the parameterized deep Q-network (PDQN) algorithm, which handles both discrete lane change actions and continuous acceleration control, enabling simultaneous lateral and longitudinal motion management.
Extensive simulations using the SUMO traffic simulator and Mosaic V2X communication framework validated PALCAS against baseline methods. The results demonstrated significant improvements in traffic efficiency (higher average speeds), driving safety (fewer collisions), comfort (smoother accelerations), destination arrival rates, and merging success rates. The federated approach also ensures scalability and privacy, making it a practical solution for real-world deployment where multiple AVs from different manufacturers must cooperate on the road.
- Uses federated reinforcement learning to coordinate multiple autonomous vehicles without sharing raw data
- Employs a priority-aware reward function based on destination urgency for both mandatory and discretionary lane changes
- Simulations show improved traffic efficiency, safety, comfort, and merging success over baseline methods
Why It Matters
Smarter, privacy-preserving lane changes could reduce accidents and traffic jams in mixed autonomous vehicle environments.