An Adaptive Framework for Manipulator Skill Reproduction in Dynamic Environments

TL;DR

This paper introduces an adaptive framework for robot skill reproduction in dynamic environments, combining novel learning, prediction, and decision-making methods to enable proactive and robust manipulation.

Contribution

It presents a new LfD representation called ELTE and integrates environment prediction with high-level decision making for proactive adaptation.

Findings

Framework achieves robust, stable behaviors in real-world tests.

Proactive adaptation outperforms reactive methods in dynamic scenarios.

ELTE effectively adapts trajectories based on predicted future states.

Abstract

Robot skill learning and execution in uncertain and dynamic environments is a challenging task. This paper proposes an adaptive framework that combines Learning from Demonstration (LfD), environment state prediction, and high-level decision making. Proactive adaptation prevents the need for reactive adaptation, which lags behind changes in the environment rather than anticipating them. We propose a novel LfD representation, Elastic-Laplacian Trajectory Editing (ELTE), which continuously adapts the trajectory shape to predictions of future states. Then, a high-level reactive system using an Unscented Kalman Filter (UKF) and Hidden Markov Model (HMM) prevents unsafe execution in the current state of the dynamic environment based on a discrete set of decisions. We first validate our LfD representation in simulation, then experimentally assess the entire framework using a legged mobile manipulator in 36 real-world scenarios. We show the effectiveness of the proposed framework under different dynamic changes in the environment. Our results show that the proposed framework produces robust and stable adaptive behaviors.