Intelligent approaches to mission planning and control for autonomous vehicles.

摘要

Autonomous vehicles are currently finding more interest in application and research. The evolution of soft computing tools, like fuzzy logic, neural networks and genetic algorithms, coincident with the emergence of high-performance, low-price commercial computers, like the 400MHz Pentium II System, and digital signal processors has provided many tools and implementation platforms to produce practical autonomous systems.; A key, but a challenging feature of autonomous vehicles is autonomy which requires high level of intelligence especially at the mission supervisory level. Task scheduling, conflict resolution, route planning, obstacle avoidance, and fault tolerance are just a few examples of high-level mission planning tasks in autonomous vehicle systems. With more complicated autonomous vehicles and more intricate missions, more intelligence is needed to fulfill the functional requirements of autonomy.; This thesis develops intelligent mission planning and control methods and approaches for large-scale fast autonomous vehicles conducting complex missions in unstructured environments. The developed approaches tackle the problem as follows: (1) An object-based world model is developed to capture the features and to support plan generation and validation for large-scale autonomous systems. (2) An A*-based global route planner is developed to generate on-line suboptimal routes with complex route's cost. Route filtering and learning support are two capabilities to support suboptimal real-time route planning. (3) A neural fuzzy local planner is developed to avoid moving obstacles and correct local deviations from the global route.; The developed approaches were implemented and integrated in the mission planning and ground station support for the Autonomous Scout Rotorcraft Testbed (ASRT) project. Towards producing a truly autonomous vehicle system, the presented approaches add the following contributions to the area of mission planning and control for AV systems: (1) The object-based world model better realizes large-scale autonomous vehicle applications. (2) The object modeling approach systematizes world modeling for planning system. (3) The presented route planning configuration allows multi-criteria global route planning. (4) The presented real-time route redirection allows on-line replanning. (5) The fast route filtering algorithm supports real-time suboptimal route planning. (6) The learning paradigm reduces route planning time for real-time autonomous vehicle applications. (7) The neural-fuzzy obstacle avoidance system allows real-time control of large-scale autonomous vehicles with multiple criteria to achieve a suboptimal avoidance maneuver to avoid a moving obstacle.