Chicago, September 14, 2014
Modern robots are expected to perform complex, unsafe, or difficult tasks. Planning and executing the motions required for these tasks is difficult due to factors such as high-dimensional configuration spaces and changing environmental conditions. Moreover, uncertainty in robot dynamics and environment makes it impossible to know ahead of time how to operate best. Recent success has been made through the integration of planning methods with tools from Machine Learning (ML). For example, clustering, reinforcement learning, and intelligent heuristics have adaptively solved planning problems in complex planning spaces, automatically identified appropriate trajectories for robots with complex dynamics, and reduced the amount of time required for planning motions.
It is the goal of this workshop to explore methods and advancements afforded by the integration of ML for the planning and execution of robot motion. Because these methods are often heuristic, issues such as safety and performance are critical. Also, learning-based questions such as problem learnability, knowledge transfer among robots, knowledge generalization, long-term autonomy, task formulation, demonstration, role of simulation, and methods for feature selection define problem solvability. We will address these issues while discussing current and future directions for intelligent planning and execution of motions for robotics systems.