Impact-aware planning and control for autonomous systems

Abstract

Autonomous systems are increasingly deployed in various unstructured domains, ranging from manufacturing and transportation to search and rescue operations. However, their effectiveness and reliability are often compromised by impacts, which can lead to unexpected state changes and even system failures. Traditional planning and control methods often lack the capability to handle impacts, resulting in suboptimal performance and adaptability. This thesis presents a novel impact-aware planning and control paradigm for autonomous systems operating in uncertain environments. Our proposed planning methods aim to address the challenges posed by abrupt state transitions caused by explicit or implicit impacts during system motion. Our planning approach is based on a trajectory optimization framework, utilizing nonlinear complementarity constraints to model motion mode switching and interactive behaviors with the environment. By explicitly considering potential impacts and incorporating them into the planning process, autonomous systems can effectively respond to unexpected events, maintain stability, and achieve mission objectives. Furthermore, the hybrid mode predictive control algorithm and data-driven passive impedance control algorithm enable autonomous systems to handle predictable and unpredictable impacts during real-time motion. The hybrid mode predictive control method extends the capabilities of model predictive control to systems with multiple motion modes, effectively addressing predictable impacts caused by tracking errors. On the other hand, the data-driven passive impedance control method enhances system stability when encountering unexpected impacts. Experimental validations on various autonomous systems demonstrate the applicability and effectiveness of the proposed impact-aware planning and control methods. The research opens up new avenues for future work in learning-based impact-aware planning and impact-aware exploration in wild environments, which further enhance the autonomy and collaboration of autonomous systems in complex and dynamic environments.

Date of Award	2024
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology
Supervisor	Shaojie SHEN (Supervisor) & Michael Yu WANG (Supervisor)