Mitigating Network Latency in Fog Robotic based Tele-Operation by Motion Segmentation and Synthesis

Nan Tian,Ajay Kummar Tanwani,Ken Goldberg,Somayeh Sojoudi

2019-01-01

Abstract:Network latency is a major problem in Cloud and Fog Robotics for human robot interaction in tele-operation. Routing delays can be highly variable in a heterogeneous computing environment, imposing challenges to reliably tele-operate a robot with a closed-loop feedback controller. In this work, we present a set of communication protocols for mitigating network latency when remotely tele-operating a robot arm. We develop a recognition, segmentation and motion synthesis approach from tele-operator demonstrations using Hidden Semi-Markov Models (HSMMs) and linear quadratic tracking controller. We evaluate the approach by first learning a motion segmentation and synthesis model for each letter in a human hand-written letter dataset, and share the model locally on both the teleoperator and the robot site. We recognize the corresponding letter from partial demonstration of the tele-operator online and synthesize the trajectory on the robot arm based on the learned models. Performance evaluation of recognition and synthesis errors in providing assistance to tele-operator while performing letter motions in the presence of communication delays suggest the feasibility of latency mitigation for Fog Robotics.