Decentralized Policy Coordination in Mobile Sensing with Consensual Communication.

Area of Science:

• Robotics
• Artificial Intelligence
• Distributed Systems

Background:

• Cooperative mobile sensing requires autonomous vehicles to navigate and cover environments.
• Decentralized navigation decisions based on local observations pose coordination challenges in dynamic environments.

Purpose of the Study:

• To propose a novel framework for cooperative mobile sensing using consensual communication in multi-agent reinforcement learning.
• To enhance coordination among decentralized vehicles by enabling information sharing.

Main Methods:

• Vehicles learn to communicate and then navigate based on received messages.
• Mutual information is used as a regularizer to promote consensus among vehicles.
• Theoretical convergence is proven under mild assumptions.

Main Results:

• The proposed algorithm is scalable and converges quickly during training.
• The algorithm significantly outperforms baseline methods in the execution phase.
• Consensual communication is shown to be crucial for coordinating decentralized vehicle behaviors.

Conclusions:

• The framework effectively addresses the challenge of coordinating decentralized vehicles in mobile sensing.
• Consensual communication enhances spatial-temporal coverage by enabling information sharing.
• The approach offers a promising direction for future research in cooperative autonomous systems.