Toward Human-Robot Teaming for Robot Navigation Using Shared Control, Digital Twin, and Self-Supervised Traversability Prediction

Collision prediction and avoidance are essential for the autonomous navigation of mobile robots. The prediction of a traversability map is a way to achieve this goal. However, this approach might fail if the prediction model is exposed to novel semantic classes unseen during self-supervised training or the environment is subject to highly dynamic motions of living and artificial entities. Included are pedestrians, animals, and vehicles that the robot can hardly handle. In this paper, we embrace this challenge by describing the development of a flexible human-robot teaming that leverages the shared control of the robot to accommodate critical situations and allow the human operator to be otherwise engaged. The operator can remotely perceive the surrounding and actively guide the robot as well as participatively share or leave full control to the robot that autonomously moves toward a desired common goal. A situation-aware control signal balances inputs from the motion planner of the autonomous robot and cognitive inputs from the remote operator issued by using affordable interfaces. The human-in-the-loop control is realized through a bidirectional wireless communication between the physical robot and its digital twin. The human-robot teaming enhances the likelihood of a safe and successful navigation under dynamic obstacles and enables a navigation without pre-defined goals. The conceptual architecture is introduced and preliminary results are shared.