An Intelligent Predictive Control Approach to the High-Speed Cross-Country Autonomous Navigation Problem

Abstract

Autonomous robot vehicles promise many ultimate civilian, military, and space applications. Off-road autonomous vehicles must engage the world exactly as they find it without relying on having it engineered to suit them. For this reason, off-road autonomous navigation is one of the most difficult automation challenges. Previous work in the area has been disappointing from the perspective of the speeds attained, and the inability of systems to travel long distances autonomously. Indeed, no system has travelled an autonomous mile or exceeded 3 m/s speeds. To date, no off-road system has approached the capabilities needed to address real applications. This thesis examines and proposes a solution to the problem of high speed autonomous navigation of outdoor vehicles. As a systems-level effort, aspects of perception, path planning, position estimation, and to a lesser extent, strategic planning and motion control are considered. the emphasis of the work has been to assess the fundamental requirements of the problem, and to validate the conclusions of this assessment through the demonstration of an improved ability to achieve a real-cross country mission on several vehicle testbeds. Results indicate the cross-country navigation systems of unprecedented capability are possible if they are designed to optimally utilize limiting computing resources. A system of unprecedented performance has been constructed and extensively tested. The essential argument of the thesis is one of architecture. An intermediate intelligent predictive control layer is introduced between the typical high-level strategic or artificial intelligence layer and the typical-level servo control layer. This new layer, the tactical layer, incorporates some deliberation, and some environmental mapping as do deliberative AI planners, yet it also emphasizes the real-time aspects of the problem as do minimalist reactive architectures. The contribution of the work is a codified systems theory that permits future design efforts to benfit from the experience and a fieldworthy prototype system that provides a baseline capability for continued research. Specific results include an analysis of the complexity of range image perception for autonomous vehicles and an associated computational image stabilization algorithm which permits highest vehicle speeds. The problem of local autonomous mobility has been formulated entirely in an optimal control context. In this context, the concepts of actuation space and hazard space replace the configuration space that is more typical of AI planners. The resulting high fidelity models stabilize coordinated control of a high speed vehicle for both obstacle avoidance and goal seeking purposes.