Learning to Optimize Edge Robotics: A Fast Integrated Perception-Motion-Communication Approach

Abstract--Edge robotics involves frequent exchanges of large-volume multi-modal data. Existing methods ignore the inter - dependency between robotic functionalities and communica tion conditions, leading to excessive communication overhead. As such, rob ots can dynamically adapt their communication strategies (i.e ., compression ratio, transmission frequency, transmit powe r) by leveraging the knowledge of robotic perception and motion d y-namics, thus reducing the need for excessive sensor data upl oads. Furthermore, by leveraging the learning to optimize (L TO) paradigm, an imitation learning neural network is designed and implemented, which reduces the computational complexi ty by over 10x compared to state-of-the art optimization solve rs. Experiments demonstrate the superiority of the proposed IP MC and the real-time execution capability of L TO. Index T erms --Edge robotics, learning to optimize. Edge robotics (ER) enables resource-constrained mobile robots to offload computation-intensive tasks to edge serve rs [1]-[5] .