Incremental Skill Learning of Stable Dynamical Systems

Abstract

Efficient skill acquisition, representation, and online adaptation to different scenarios has become of fundamental importance for assistive robotic applications. In the past decade, dynamical systems (DS) have arisen as a flexible and robust tool to represent learned skills and to generate motion trajectories. This work presents a novel approach to incrementally modify the dynamics of a generic autonomous DS when new demonstrations of a task are provided. A control input is learned from demonstrations to modify the trajectory of the system while preserving the stability properties of the reshaped DS. Learning is performed incrementally through Gaussian process regression, increasing the robot's knowledge of the skill every time a new demonstration is provided. The effectiveness of the proposed approach is demonstrated with experiments on a publicly available dataset of complex motions.

Merging Position and orientation Motion Primitives

“…Dynamical systems are used to plan in joint or Cartesian space, and, in Cartesian space, to encode both position and orientation [13]- [16]. Moreover, robots driven by stable systems are able to reproduce complex paths [7]- [10], to incrementally update a predefined skill [11], [12], and to avoid possible collisions [17]- [21].…”

Combining decision making and dynamical systems for monitoring and executing manipulation tasks

“…DS trajectories are generated at runtime, allowing for online motion replanning to handle unexpected perturbations [5][6][7][8]. A stable dynamical system can be learned from human demonstrations [9] in an incremental way [10,11]. Finally, suitable control can be applied to constrain the motion within a certain region and adapt to changes in the workspace [12].…”

An Energy-based Approach to Ensure the Stability of Learned Dynamical Systems

“…DS can on-line replan the robot's trajectory to cope with changes in the target position or unexpected obstacles [10]- [13]. DS motions can be incrementally updated as novel task demonstrations are provided [14], [15] or confined to constrained domains [16] without compromising the convergence to the goal. Finally, DS have been used to encode primitive robotic skills or movement primitives [17], which can be properly scheduled to execute complex tasks [18], [19].…”