Traction Awareness and Control Through Haptic Teleoperation of Space Rovers

Abstract

Bilateral teleoperation systems have demonstrated significant potential for addressing travel reduction and path deviation in wheeled rover operation on soft terrains. However, their application to planetary exploration, such as tele-driving rovers on complex lunar terrains under interplanetary communication delays and poor lighting conditions remains challenging. Thus, significant research advancements are necessary to achieve a high-fidelity teleoperation system with enhanced situational awareness and improved tracking performance. This dissertation addresses four research challenges from the literature: (1) lack of a high-fidelity simulation method for traction analysis of the lug wheel on soft terrains, (2) limitations of existing control schemes in achieving a fair trade-off between stability and performance for closed-loop bilateral teleoperation systems, (3) the unavailability of adaptive and generalized deep learning-based predictor approaches to compensate for large delays in unknown trajectory in the teleoperation systems, (4) the inadequacy of predictive displays in providing reliable visual feedback under poor lighting to compensate for delays and the absence of external viewpoint. This dissertation aims to design a robust and stable bilateral teleoperation system that facilitates effective remote control of lunar rovers traversing complex terrains under large delays and lighting constraints. Initially, a high-fidelity simulation method is developed and validated to model the wheel-terrain interaction dynamics, focusing on lug wheel on soft terrain. The proposed method utilized a modified terramechanics model integrated with the lug dynamics effect. Experimental validation demonstrated the method's high accuracy, establishing its reliability for simulating a lunar rover virtual environment. Subsequently, a human-in-the-loop test platform is developed based on this method, enabling the analysis of closed-loop stability and transparency of bilateral teleoperation systems for rovers under slippage and large delays. A new first-order time-delayed (FOTD) predictor-based control scheme is proposed, effectively stabilizing the system without compromising performance, thus outperforming existing control schemes. Although performance analysis demonstrated moderate delay compensation, the overall fidelity of the teleoperation system remained inadequate. To address this, a deep learning (DL), physics-informed long short-term memory (PiLSTM)-based, predictor framework is developed to improve adaptability and generalizability for better delay compensation, resulting in improved teleoperation system fidelity. By integrating the LSTM network with first-order time-delayed dynamics, the PiLSTM predictor demonstrates greater robustness to prediction errors. Compared with the FOTD predictor, the PiLSTM approach achieves a 29.7% improvement in system fidelity, leading to enhanced situational awareness and tracking performance. Finally, a frame-based predictive display method (PDM) is developed using the DL predictor to compensate for visual feedback delays and counteract the absence of the external view. In this approach, predicted states are superimposed on the delayed visual feed in real-time, yielding an augmented display view. This improved maneuverability and reduced workload.

Date of Award	22 May 2025
Original language	American English
Supervisor	Seneviratne Seneviratne (Supervisor)