Benefits of Low-Cost Bio-Inspiration in the Age of Overparametrization

arXiv cs.RO / 4/23/2026

📰 NewsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The study investigates how controller learning behaves when the parameter space is large, focusing on two bio-inspired/MLP paradigms (central pattern generators and multi-layer perceptrons) for robot control.
It finds that when task input/output spaces are small and performance is bounded, increasing model depth or parameter count can hinder learning rather than improve it.
Across controller optimization experiments using evolutionary and reinforcement trainers, shallow MLPs and densely connected CPGs outperform deeper MLPs and Actor-Critic-style architectures.
The authors introduce a “Parameter Impact” metric showing that reinforcement methods often require more additional parameters without corresponding gains in performance, which favors evolutionary strategies.

Abstract

While Central Pattern Generators (CPGs) and Multi-Layer Perceptrons (MLP) are widely used paradigms in robot control, few systematic studies have been performed on the relative merits of large parameter spaces. In contexts where input and output spaces are small and performance is bounded, having more parameters to optimize may actively hinder the learning process instead of empowering it. To empirically measure this, we submit a given robot morphology, with limited proprioceptive capabilities, to controller optimization under two bio-inspired paradigms (CPGs and MLPs) with evolutionary- and reinforcement- trainer protocols. By varying parameter spaces across multiple reward functions, we observe that shallow MLPs and densely connected CPGs result in better performance when compared to deeper MLPs or Actor-Critic architectures. To account for the relationship between said performance and the number of parameters, we introduce a Parameter Impact metric which demonstrates that the additional parameters required by the reinforcement technique do not translate into better performance, thus favouring evolutionary strategies.