FLDQN: Cooperative Multi-Agent Federated Reinforcement Learning for Solving Travel Time Minimization Problems in...

Area of Science:

• Artificial Intelligence
• Transportation Engineering
• Distributed Systems

Background:

• Increasing traffic volume leads to congestion, pollution, and longer travel times.
• Deep Reinforcement Learning (DRL) shows promise in traffic management but is limited to single-agent systems.
• Cooperative multi-agent reinforcement learning (MARL) is challenging due to agent management and collaboration complexities.

Purpose of the Study:

• To introduce a cooperative multi-agent federated reinforcement learning algorithm (FLDQN) for optimizing road network utilization.
• To address the challenge of agent cooperation in dynamic MARL scenarios.
• To minimize travel times and reduce traffic congestion.

Main Methods:

• Developed FLDQN, a federated reinforcement learning algorithm for cooperative MARL.
• Utilized the SUMO simulator for multi-agent interactions and environment modeling.
• Agents employed deep Q-learning, sharing model updates via a federated server for collective policy enhancement.

Main Results:

• FLDQN achieved an average reduction of over 34.6% in travel time compared to non-cooperative methods.
• Demonstrated significant reduction in traffic congestion.
• Lowered computational overhead through distributed learning.

Conclusions:

• Agent cooperation is crucial for effective traffic management in multi-agent systems.
• FLDQN offers an innovative solution for enabling cooperation and knowledge sharing among intelligent agents.
• Federated learning facilitates enhanced policy learning by leveraging collective experiences in dynamic environments.