RSS 2015 Workshop

Bridging the Gap between Data-driven and Analytical Physics-based Grasping and Manipulation

July 16, Rome

"Towards Machine Learning of Grasping and Manipulation", Jan Peters, TU Darmstadt

In the future, robots could be used to take care of the elderly, perform household chores, and assist in hazardous situations. However, such applications require robots to manipulate objects in unstructured and everyday environments. Hence, in order to perform a wide range of tasks, robots will need to learn versatile manipulation skills that generalize between different scenarios and objects. In this talk, I will present methods developed for robots to learn versatile manipulation skills. The robot can first decompose a task into common subtasks by segmenting the task into phases. Specific skills are then learned for each sub-task. These skills can initially be acquired using imitation learning and, subsequently, mastered through trial-and-error using reinforcement learning methods. By learning from multiple objects and scenarios, the robot can learn versatile skills that generalize to new situations.