Improved Performance for Multi-Arm Harvest Robot System with Edge Computing and Distributed Communication

Abstract

It is of practical significance to use robots for fruit harvesting instead of labor. According to the increased complexity and computing power for multi-arm harvest robots, a robotic system with edge computing and distributed communication is proposed to ensure performance and control costs. In this system, NVIDIA Jetson TX2 is used for object prediction to share the computation of the host platform. Then, in order to ensure real-time performance of the vision framework, a mode of linking between multiple cameras and the platforms of the host and edge is designed, which allocates tasks for visual processing to each platform and centralizes images and labels to the host via distributed communication based on ROS topics. Finally, we verified this approach that can increase the speed of prediction and ensure the fluency of depth streams, improving the performance of the system and obtaining a more efficient harvest.