Logic Learning from Demonstrations for Multi-step Manipulation Tasks in Dynamic Environments

TL;DR

This paper introduces Logic-DMP, a novel framework combining task and motion planning with optimal control to enable robots to learn, generalize, and react to disturbances in complex, long-horizon manipulation tasks within dynamic environments.

Contribution

The paper presents Logic-DMP, integrating TAMP with DMP for improved generalization and reactivity in long-horizon manipulation tasks under dynamic conditions.

Findings

Logic-DMP demonstrates fast generalization to task variants.

Logic-DMP effectively handles disturbances in dynamic environments.

Experimental results show superior reactivity compared to baselines.

Abstract

Learning from Demonstration (LfD) stands as an efficient framework for imparting human-like skills to robots. Nevertheless, designing an LfD framework capable of seamlessly imitating, generalizing, and reacting to disturbances for long-horizon manipulation tasks in dynamic environments remains a challenge. To tackle this challenge, we present Logic Dynamic Movement Primitives (Logic-DMP), which combines Task and Motion Planning (TAMP) with an optimal control formulation of DMP, allowing us to incorporate motion-level via-point specifications and to handle task-level variations or disturbances in dynamic environments. We conduct a comparative analysis of our proposed approach against several baselines, evaluating its generalization ability and reactivity across three long-horizon manipulation tasks. Our experiment demonstrates the fast generalization and reactivity of Logic-DMP for handling task-level variants and disturbances in long-horizon manipulation tasks.