LAMPS

Abstract

Learning from Demonstrations (LfD) methods are applied to transfer human skills to robots from expert demonstrations, enabling them to perform complex tasks. However, existing methods often struggle to handle such long-horizon human skills as cleaning or wiping stains on the surface, which involve multiple periodic and transitional movement primitives. To address this limitation, this paper first proposes a novel framework for segmenting, learning, and generalizing multi-periodic human skills, enabling robots to effectively learn different movement primitives and execute these skills in new scenarios. Specifically, The framework introduces an unsupervised learning method to segment long-horizon human demonstrations into periodic and discrete movement primitives. Further, a novel type of discrete dynamical movement primitives, namely transitional movement primitives, is employed to enhance the fluidity of combining different periodic movement primitives in skills. To validate the effectiveness of the proposed approach, we conduct extensive experimental evaluations, including step-by-step validation of each procedure's methods in simulation and the implementation of the entire presented framework in the real world. The results confirm that the proposed framework accurately learns and generalizes multi-periodic human skills, providing a feasible solution for transferring complex multi-periodic demonstrations to robots in practical applications.

Main Contribution

We introduce LAMPS💡, a novel robot skill learning framework consisting of movement primitives (MPs) segmenting, merging and generalizing. The main contributions of our work are summarized as follows:

The multi-periodic human skills are modeled for the first time by merging periodic MPs and discrete MPs, successfully providing detailed descriptions of such kinds of widely-used tasks and addressing a gap in the existing literature.
The proposed framework is constructed with the knowledge in unsupervised learning and LfD, which enables fluent generalization of MPs while preserving the dynamic features of demonstrations, enhancing both validity and efficiency.
The efficacy of the proposed framework is evaluated through both simulations and experiments, which validate the framework's efficacy in learning and generalizing multi-periodic skills, demonstrating robust performance and adaptability.

Here we show a brief overview of the proposed framework for Learning Adaptive Multi-Periodic Skills: LAMPS💡

Methodology and Evaluation

An trajectory segmentation process based on the unsupervised learning method OPTICS (Ordering Points To Identify the Clustering Structure) is proposed and used to effectively segment different MPs from long human demonstrations.

The transitional MPs are introduced to combine with different periodic MPs in a fluent way.

The proposed transitional MP method is evaluated to combine with periodic MPs in different phases, with different positions, velocities, and accelerations.

Real-World Experiments

We executed and estimate one of the typical multi-periodic movement skills, wiping the whiteboard to evaluate the proposed framework LAMPS.

Human Skill of Wiping Whiteboard

Experimental Scene diagram

Experimental Result Diagram

Here we show a clear overview of the experimental procedure using the propose framework. Please visit here to see more details about our work experiment part.

Example Results Analysis

Several periodic MPs and transitional MPs are contained in multi-periodic skills in human daily life. The following results show the fluent performance