Risk-Aware Virtual-Reality Teleoperation with Gaze-Conditioned Control for High-Temperature Casting

Abstract

High-temperature casting exposes on-site workers to irreversible thermal hazards, where robotic teleoperation provides an effective hazard-isolation approach. Traditional 2D teleoperation suffers from insufficient spatial perception, while existing virtual reality (VR) systems often adopt an open-loop "input-to-execute" approach, which relies on operator inputs aligning with visual attention. Meanwhile, geometric-only VR models cannot convey thermo-structural coupling risks of casting end-effectors, leaving operators with challenges to anticipate deformation and spill hazards. This paper develops a dual-loop safety-oriented framework for VR casting teleoperation: it uses eye tracking to take attention validity as a precondition for command execution, and maps thermo-structural finite element analysis (FEA) results into runtime risk feedback and control constraints. Two within-subject controlled experiments validate the framework: the gaze-conditioned module reduces misoperations by 44.62% with a 5.08% task time increase; the FEA-driven module reduces boundary violations by 42.83%, safety violation frequency by 46.21%, and end-effector acceleration fluctuation by 22.83%, with a 6.52% time increase.