Gaze-Guided Semi-Autonomous Quadruped Robot for Enhanced Assisted Living

Abstract

Robotic automation represents mankind's next leap. While the industry has already embraced robotization, robot-driven domestic aid is only at the beginning of the revolution. Dealing with unconstrained daily environments is more challenging than automating manufacturing production lines. The growing diffusion of semi-autonomous assistive quadrupedal robots that can handle obstacles triggered the change. Nonetheless, robot control still requires active human supervision, which can be tedious in ordinary surroundings or even impossible in clinical circumstances. This paper tackles the robot control challenge and introduces the deployment of gaze-controlled semi-autonomous assistive quadrupedal robots. We focus on noninvasive gaze-tracking performed on smart glasses to guide a semi-autonomous assistive quadrupedal robot toward the user's focused remote object. Specifically, we propose (i) a preliminary setup to exploit the potential of gaze-tracking in quadrupedal robot control with the prospective of enabling delocalized object grasping and (ii) an assessment of two state-of-the-art image-matching algorithms, considering both accuracy and power footprint. Results show that the proposed solution enables piloting of quadrupedal robots toward a target highlighted by the user gaze with an accuracy of less than 20 cm and an inference time of ~ 200 ms running the algorithms on an on-board embedded computational unit.