Efficient and Diverse Generative Robot Designs using Evolution and Intrinsic Motivation

TL;DR

This paper tackles the computational bottleneck and limited diversity of morpho-evolution with learning (MEL) for robot design. It introduces MEHK, which uses intrinsic motivation via homeokinesis to rapidly explore and evaluate a large morpho-spatial design space without lengthy per-design learning, implemented as asynchronous morpho-evolution with a homeokinetic controller. Empirical results show MEHK yields higher exploration scores, greater design diversity, and superior downstream task performance compared to a baseline morpho-evolution with a fixed controller, while using substantially fewer computational resources. The approach offers a practical, scalable path to generating viable and varied robot bodies and controllers for complex tasks in dynamic environments.

Abstract

Methods for generative design of robot physical configurations can automatically find optimal and innovative solutions for challenging tasks in complex environments. The vast search-space includes the physical design-space and the controller parameter-space, making it a challenging problem in machine learning and optimisation in general. Evolutionary algorithms (EAs) have shown promising results in generating robot designs via gradient-free optimisation. Morpho-evolution with learning (MEL) uses EAs to concurrently generate robot designs and learn the optimal parameters of the controllers. Two main issues prevent MEL from scaling to higher complexity tasks: computational cost and premature convergence to sub-optimal designs. To address these issues, we propose combining morpho-evolution with intrinsic motivations. Intrinsically motivated behaviour arises from embodiment and simple learning rules without external guidance. We use a homeokinetic controller that generates exploratory behaviour in a few seconds with reduced knowledge of the robot's design. Homeokinesis replaces costly learning phases, reducing computational time and favouring diversity, preventing premature convergence. We compare our approach with current MEL methods in several downstream tasks. The generated designs score higher in all the tasks, are more diverse, and are quickly generated compared to morpho-evolution with static parameters.

Figures (9)

• Figure 1: Robots generated by our approach MEHK. Exploration values refer to the evaluation phase during morpho-evolution. The other values correspond to the downstream task scores.
• Figure 2: Asynchronous Morpho-Evolution. Our approach, MEHK, uses intrinsically motivated behaviour, homoekinesis, for evaluation.
• Figure 3: Five components for the robot's design.
• Figure 4: a) Arena used for the evaluation of the robots during MEHK generative phase. b), c) and d), are the downstream tasks environments. For clarity, d) only shows the beginning of the arena.
• Figure 5: Comparison of MEHK (blue) and MEFC (orange). On the left, the total computational time and on the right, best exploration score obtained from each replicate.