Multi-expert learning of adaptive legged locomotion

TL;DR

This paper introduces MELA, a multi-expert learning framework that synthesizes adaptive locomotion skills for quadruped robots by combining pre-trained experts with a gating network, enabling real-time adaptation to new situations.

Contribution

The paper presents a novel multi-expert learning architecture that dynamically blends expert skills to produce adaptable locomotion behaviors in robots.

Findings

Successful real-time adaptive locomotion on a quadruped robot

Effective combination of pre-trained experts for diverse skills

Demonstrated adaptation to unseen scenarios

Abstract

Achieving versatile robot locomotion requires motor skills which can adapt to previously unseen situations. We propose a Multi-Expert Learning Architecture (MELA) that learns to generate adaptive skills from a group of representative expert skills. During training, MELA is first initialised by a distinct set of pre-trained experts, each in a separate deep neural network (DNN). Then by learning the combination of these DNNs using a Gating Neural Network (GNN), MELA can acquire more specialised experts and transitional skills across various locomotion modes. During runtime, MELA constantly blends multiple DNNs and dynamically synthesises a new DNN to produce adaptive behaviours in response to changing situations. This approach leverages the advantages of trained expert skills and the fast online synthesis of adaptive policies to generate responsive motor skills during the changing tasks. Using a unified MELA framework, we demonstrated successful multi-skill locomotion on a real quadruped robot that performed coherent trotting, steering, and fall recovery autonomously, and showed the merit of multi-expert learning generating behaviours which can adapt to unseen scenarios.