Real-Time Joint Trajectory Optimization and Predictive Safety Filtering for Telepresence Avatar Robotics

Robotic telepresence allows humanity to remotely conduct critical life-saving physical work or connect with faraway relatives. However, achieving this with an intuitive, general-purpose avatar agent demands seamless real-time motion retargeting; hence, inaccurate, sluggish robotic movement is unusable. In this paper, a new trajectory controller for the telepresence humanoid Avatar-Hubo is developed to refine manipulator accuracy, responsiveness, and safety. By using the proposed filter fusion-based trajectory generator, Avatar-Hubo's average joint motion latency decreased 90%, dropping from 1.5
$s$
to 0.15
$s$
. Additionally, end-effector tracking accuracy Root Mean Square Error improved by 97 % on average. Furthermore, the risk of dangerous robot motions is remedied via the proposed predictive safety filter that identifies and prevents motions requiring excessive speeds. Ultimately, the new controller possessed superior real-time motion performance, aiding in general robotic usage and inducing the sensation of telepresence.