Bio-Inspired Plastic Neural Networks for Zero-Shot Out-of-Distribution Generalization in Complex Animal-Inspired Robots

2025-03-16

AI Summary

Artificial neural networks often fail under out-of-distribution (OOD) conditions—such as unfamiliar terrain or limb damage—due to poor generalization and weight divergence in conventional learning rules. Method: This paper proposes a biologically inspired plastic neural network with dynamic weight normalization, integrating local Hebbian synaptic plasticity, principal-component-based weight analysis, and an end-to-end locomotion control architecture to stabilize learning and prevent unbounded weight growth. Contribution/Results: To our knowledge, this is the first work to achieve zero-shot sim-to-real transfer and OOD robust generalization on high-dimensional legged robots—a 18-DOF dung-beetle–inspired platform and a 16-DOF gecko-inspired robot—without online fine-tuning. The method autonomously adapts to unseen challenging conditions, including uneven terrain and joint failures. Experiments demonstrate significant improvements in policy robustness and environmental generalization over baseline approaches, establishing a novel paradigm for autonomous, adaptive locomotion in open, dynamic environments.

Technology Category

• Intelligent Robots: Behavior Learning & Control
• Machine Learning: Bio-inspired Learning
• Natural Language Processing: Safety and Robustness

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Artificial neural networks can be used to solve a variety of robotic tasks. However, they risk failing catastrophically when faced with out-of-distribution (OOD) situations. Several approaches have employed a type of synaptic plasticity known as Hebbian learning that can dynamically adjust weights based on local neural activities. Research has shown that synaptic plasticity can make policies more robust and help them adapt to unforeseen changes in the environment. However, networks augmented with Hebbian learning can lead to weight divergence, resulting in network instability. Furthermore, such Hebbian networks have not yet been applied to solve legged locomotion in complex real robots with many degrees of freedom. In this work, we improve the Hebbian network with a weight normalization mechanism for preventing weight divergence, analyze the principal components of the Hebbian's weights, and perform a thorough evaluation of network performance in locomotion control for real 18-DOF dung beetle-like and 16-DOF gecko-like robots. We find that the Hebbian-based plastic network can execute zero-shot sim-to-real adaptation locomotion and generalize to unseen conditions, such as uneven terrain and morphological damage.

Problem

• Enhance neural networks for zero-shot OOD generalization
• Prevent weight divergence in Hebbian learning networks
• Apply plastic networks to complex legged robot locomotion

Innovation

• Enhanced Hebbian learning with weight normalization
• Principal component analysis of Hebbian weights
• Zero-shot sim-to-real adaptation for robot locomotion