Rapid co-design of Buoyancy-assisted robots for Challenging Locomotion using Gaussian Evolutionary Specialists

2026-06-05

AI Summary

Legged robots often face limitations in morphology-control co-optimization due to high computational costs and insufficient behavioral diversity. This work proposes the Gaussian Evolutionary Specialists (GES) framework, which decouples design space partitioning from policy learning by evolving Gaussian regions that allocate specialized policies and embedding this process within a design sampling loop to enable efficient co-design. GES explicitly models diverse locomotion behaviors, thereby avoiding performance collapse commonly observed in monolithic policies or mixture-of-experts approaches. Evaluated on the BALLU robot, GES improves performance by 5–25% over a universal policy, achieves a threefold increase in obstacle traversal height—reaching 24 cm in physical experiments—and reduces design optimization time by 37%.

Abstract

Designing high-performance legged robots requires jointly optimizing morphology and control. Model-free Reinforcement Learning (RL) offers an alternative to model-predictive control for developing robust controllers without explicitly specifying robot dynamics. Thus, we have seen theuse of RL to train controllers and evaluate designs for robot morphology optimization. While RL has shown success inlocomotion, using it in the co-design inner loop is expensive due to repeated policy training. Universal policies conditioned on morphology offer a promising alternative, but suffer from behavioral diversity collapse, converging to a single strategy that performs sub-optimally across designs. On the other hand, end-to-end Mixture-of-Experts (MoE) architectures fail due to a collapse in its representation. We propose Gaussian Evolutionary Specialists (GES), a framework that decouples design-space partitioning from policy learning to capture diverse behaviors explicitly. GES assigns specialist policies to evolving Gaussian regions and iteratively refines them via training, probing, and territory expansion. The resulting specialists are integrated into a design sampling loop, replacing costly re-training with direct evaluation. When tested on the Buoyancy-Assisted Light Legged Unit (BALLU), GES discovers designs with 5 - 25% higher performance than naive universal policies. On hardware, a GES optimized design overcomes a 24 cm tall obstacle - 3x improvement over the baseline BALLU design. Moreover, GES curtails design optimization time by 37%.

Problem

• co-design
• locomotion
• behavioral diversity
• morphology optimization
• reinforcement learning

Innovation

• Gaussian Evolutionary Specialists
• robot co-design
• morphology optimization
• reinforcement learning
• Mixture-of-Experts