Information-Centric Networking for Intelligent Cooperative Perception of Air Vehicles

In this project, we propose to develop a distributed deep reinforcement learning-based intelligent cooperative perception system (iCOOPER) with information-centric networking, which enables autonomous air vehicles (AAVs) to share sensory information with other AAVs and infrastructure for extending view coverage, removing blind spots, and enhancing object detection accuracy. Specifically, i) a sensory data communication framework along with communication protocols based on information-centric networking (ICN) will be developed, which allows AAVs to flexibly name, publish, request and retrieve/subscribe data for cooperative perception according to the sensed regions with multiple resolutions. ii) A deep reinforcement learning-based adaptive transmission scheme will be designed, which dynamically select the data and compression format to obtain based on the importance and wireless network state for optimizing the cooperative perception in terms of latency and object detection accuracy. iii) An efficient real-time 3D point cloud stream compression scheme based on recurrent neural networks will be implemented to reduce network load and delay, while maintaining accurate cooperative perception. iv) An point cloud fusion scheme will be developed, which can compensate network latency and accurately fuse the point clouds from different AAVs. The proposed system will be integrated and evaluated through analysis, simulation, and prototyping to demonstrate its feasibility.