Genesis: Advancing Towards Efficient Embodiment Co-Design

Abstract

Embodiment co-design aims to optimize a robot's morphology and control simultaneously. Previous research has demonstrated its potential for generating environment-adaptive robots. However, the problem is inherently combinatorial and the morphology is changeable and agnostic in its vast search space, optimization efficiency remains a hard nut to crack. We prove that the inefficient morphology representation and unbalanced reward signals between the design and control stages are key obstacles against efficiency. In order to advance towards efficient embodiment co-design to unlock its full potential, we propose Genesis, which utilizes (1) a novel topology-aware self-attention architecture, enabling efficient morphology representation while enjoying lightweight model sizes; (2) a temporal credit assignment mechanism for co-design that ensures balanced reward signals for optimization. With our simple-yet-efficient methods, Genesis achieves average 60.52% performance improvement against the strongest baselines.

Demo: Random Embodied Agents generated by our embodiment co-design approach, Genesis .

Methods

Genesis is an RL-based framework for embodiment co-design, which optimizes morphology and control simultaneously. It achieves centralized, zero-decay message processing based on limb-level self-attention, message localization via Topology Position Encoding, and enhanced temporal credit assignment for balanced reward signals.

For embodiment co-design, besides differentiating message source within the body, we require an evolution-aware morphology representation. We propose Topology Position Encoding (TopoPE), to better adapt to the evolving process. It not only achieves message localization but also faciliates knowledge sharing within the dynamically growing body.

Visualization

We randomly pick embodied agents generated by Genesis for visualization.

Crawler

TerrainCrosser

Cheetah

Swimmer

Glider-Regular

Glider-Medium

Glider-Hard

Walker-Regular

Walker-Medium

Walker-Hard

Experiments

Experiments show that Genesis achieves an average 60.52% performance improvement over the strongest baselines. Notably, each model generated by Genesis is only 10-20 MB in size, but has already exhibited exciting capabilities in morphology evolution and motion control, demonstrating promising potential for scaling-up towards more complex embodied agents and embodiment systems in the expected future.