Vehicle teleoperation has the ability to bridge the gap between completely automated driving and manual driving by remotely monitoring and operating autonomous vehicles when their automation fails. Among many challenges related to vehicle teleoperation, the considered ones in this work are variable time delay, saturation of actuators installed in vehicle, and environmental disturbance, which together limit the teleoperation performance. State-of-the-art predictive techniques estimate vehicle states to compensate for the delays, but the predictive states do not account for sudden disturbances that the vehicle observes, which makes the human-picked steer inadequate. This inadequacy of steer deteriorates the path-tracking performance of vehicle teleoperation. In the proposed successive reference-pose-tracking (SRPT) approach, instead of transmitting steering commands, the reference trajectory, in the form of successive reference poses, is transmitted to the vehicle. This paper introduces a method of generation of successive reference poses with a joystick steering wheel and compares the human-in-loop path-tracking performance of the Smith predictor and SRPT approach. Human-in-loop experiments (with 18 different drivers) are conducted using a simulation environment that consists of the integration of a real-time 14-DOF Simulink vehicle model and Unity game engine in the presence of bidirectional variable delays. Scenarios for performance comparison are low adhesion ground, strong lateral wind, tight corners, and sudden obstacle avoidance. Result shows significant improvement in reference tracking and in reducing human effort in all scenarios using the SRPT approach.
Keywords: NMPC; Smith predictor; latency; successive reference-pose tracking; time delay; vehicle teleoperation.