Multiform Adaptive Robot Skill Learning from Humans

TL;DR

This paper introduces a multiform adaptive learning approach for robotic manipulation, enabling robots to handle complex objects with mixed rigidity and softness, enhancing human-robot interaction capabilities.

Contribution

It presents a novel multiform learning framework that combines demonstration, definition, and evaluation, allowing robots to learn handling partly rigid and soft objects with advanced contact control.

Findings

Robots can learn to manipulate partly rigid and soft objects.

The approach improves contact control in manipulation tasks.

Enhanced capabilities for human-robot interaction.

Abstract

Object manipulation is a basic element in everyday human lives. Robotic manipulation has progressed from maneuvering single-rigid-body objects with firm grasping to maneuvering soft objects and handling contact-rich actions. Meanwhile, technologies such as robot learning from demonstration have enabled humans to intuitively train robots. This paper discusses a new level of robotic learning-based manipulation. In contrast to the single form of learning from demonstration, we propose a multiform learning approach that integrates additional forms of skill acquisition, including adaptive learning from definition and evaluation. Moreover, going beyond state-of-the-art technologies of handling purely rigid or soft objects in a pseudo-static manner, our work allows robots to learn to handle partly rigid partly soft objects with time-critical skills and sophisticated contact control. Such capability of robotic manipulation offers a variety of new possibilities in human-robot interaction.