Experiments in robot learning

Abstract

The authors attempt to bridge the planning gap problem that exists between the AI (artificial intelligence) and robotics communities. the objective was determining how procedural and parameter knowledge could be represented and used in task planning. Because AI planners lack the ability to deal with kinematic and kinetic information of a world model, and robot planners possess poor reasoning ability, an advanced robotics research environment was considered the appropriate demonstrator. Experiments were conducted on two generic operations that are common to robotics work, grasping and pushing an object, working from a starting point that was randomly selected. Task parameter constraints are derived from rules induced from the small amounts of raw sensory data collected. These rules are then used to indicate whether a task, such as object grasp or push, could be successfully completed.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>