Goal driven integrated intelligence for autonomous systems intelligence planifiée pour les systèmes autonomes

点击翻译按钮获取中文摘要

Developing autonomous systems--whether hardware (e.g., robots) or software (e.g., computer-aided education tools)--that can handle novel situations for which they have not been explicitly programmed remains a complex and challenging task. Most of the current tools and methodologies are, at best, in the experimental stage. Reacting to unanticipated events, planning and monitoring the steps to accomplish a task, and learning from experience all remain substantial problems that can be further complicated by the system's intrinsic limitations in perceiving or acting on its environment. One objective of this project is to develop new automated planning algorithms for complex robotics processes that are beyond the capability of today's planning systems and to integrate these algorithms with other robot decision-making processes. This will enable robots to accomplish tasks that are difficult with current approaches, such as assisting people in everyday life activities, performing surgery or inspecting and repairing space stations. Another objective is to develop new planning algorithms for tutorial dialogues that are impossible with current approaches and to integrate these algorithms into simulators to support the learning and control of robot manipulation procedures. This will increase the safety of robot manipulations and reduce training costs (simulations are less expensive than the actual robotics equipment and there is no risk of injury or equipment damage). These simulators can be installed anywhere and anytime to provide "just-in-time" training to support continuing education necessary for operators to maintain their skill levels (e.g. during long missions). This project primarily targets the application domain of robotics, but the research results have potential applications in many others.