Growable and interpretable neural control with online continual learning for autonomous lifelong locomotion learning machines

2025-05-15

The international journal of robotics research

AI Summary

Addressing four key challenges in autonomous lifelong motor learning—lack of interpretability, low sample efficiency, insufficient knowledge reuse, and catastrophic forgetting—this paper proposes GOLLUM, a *growable* and *interpretable* online continual learning framework. Methodologically, GOLLUM features a hierarchical-columnar dual-interpretable architecture: an upper layer enables neurogenesis-inspired unsupervised skill emergence, while a lower layer employs a ring-structured columnar network coupled with a skill-encoding mapping layer to support cross-skill parameter transfer and compositional reinforcement. Its primary contribution is the first demonstration on a physical hexapod robot of fully autonomous, zero-human-intervention, sparse-reward-driven acquisition and compositional generalization of multi-gait locomotion (walking, incline climbing, jumping). Skills are acquired within one hour; compositional mechanisms significantly accelerate learning of novel tasks. Real-world hardware evaluation confirms complete mitigation of catastrophic forgetting, demonstrating strong robustness and genuine lifelong adaptability.

Technology Category

• Machine Learning: Life-Long and Continual Learning
• Intelligent Robots: Learning & Optimization for ROB
• Search and Optimization: Learning to Search

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Architectures and workflows that use LLMs for crowd work

Abstract

Continual locomotion learning faces four challenges: incomprehensibility, sample inefficiency, lack of knowledge exploitation, and catastrophic forgetting. Thus, this work introduces growable online locomotion learning under multicondition (GOLLUM), which exploits the interpretability feature to address the aforementioned challenges. GOLLUM has two dimensions of interpretability: layer-wise interpretability for neural control function encoding and column-wise interpretability for robot skill encoding. With this interpretable control structure, GOLLUM utilizes neurogenesis to unsupervisely increment columns (ring-like networks); each column is trained separately to encode and maintain a specific primary robot skill. GOLLUM also transfers the parameters to new skills and supplements the learned combination of acquired skills through another neural mapping layer added (layer-wise) with online supplementary learning. On a physical hexapod robot, GOLLUM successfully acquired multiple locomotion skills (e.g., walking, slope climbing, and bouncing) autonomously and continuously within an hour using a simple reward function. Furthermore, it demonstrated the capability of combining previous learned skills to facilitate the learning process of new skills while preventing catastrophic forgetting. Compared to state-of-the-art locomotion learning approaches, GOLLUM is the only approach that addresses the four challenges above mentioned without human intervention. It also emphasizes the potential exploitation of interpretability to achieve autonomous lifelong learning machines.

Problem

• Addresses incomprehensibility in continual locomotion learning
• Solves sample inefficiency and catastrophic forgetting issues
• Enables autonomous skill acquisition and combination

Innovation

• Interpretable neural control for lifelong locomotion learning
• Unsupervised column growth for skill encoding
• Online supplementary learning for skill transfer