A Roadmap for Research in Robot Planning

The next generations of autonomous robots will be characterized by being more general than the current one — in at least three respects. First, the robots will be able to successfully carry out multiple, diverse, and possibly interfering tasks in changing and partly unknown environments. Second, they will be able to improve their performance by autonomously adapting their control software for the kinds of tasks they are given and the environments they are to operate in. Third, they will be able to perform novel tasks without learning the achievement of these tasks in long and tedious learning sessions. As a research community, we believe that these aspects of generality cannot be achieved without the robots being capable of planning their course of action based on foresight and without them being able to autonomously learn better control routines. To achieve more generality in these respects, parts of the control programs, called the robot plans, are to be represented explicitly, such that the robot can reason about them and revise them. This approach to autonomous robot control is called robot planning or — more generally — plan-based control. The objective is to use plans and the respective reasoning mechanisms for them as resources for increasing the generality and the performance of robot control programs. In this research road map we show and analyze the potential for impact of robot planning on autonomous robot and agent control. Our goal is to show appropriate ways of how this potential can be realized, and lay down what the research community, application developers, and funding agencies can do to stimulate and accelerate scientific and technological breakthroughs. We propose a set of challenge application scenarios and a technological milestones for near-term, middle-term, and long-term research and technology development projects that must be achieved to control autonomous robots over extended periods of time, in natural and unmodified human environments, and to perform complex jobs with them. We begin the roadmap by defining the domain of robot planning and its relation to the related fields of control theory, autonomous agents, and artificial intelligence. We then propose a framework of computational mechanisms that are necessary and make successful plan-based robot control possible. Thereafter, we illustrate the state of the art by sketching several autonomous robots that employ plan-based control mechanisms and briefly review current technological developments of the respective computational mechanisms. Subsequently, we examine a variety of challenge demonstration scenarios …