Abstract:The automatic design of robots has existed for 30 years but has been constricted by serial non-differentiable design evaluations, premature convergence to simple bodies or clumsy behaviors, and a lack of sim2real transfer to physical machines. Thus, here we employ massively-parallel differentiable simulations to rapidly and simultaneously optimize individual neural control of behavior across a large population of candidate body plans and return a fitness score for each design based on the performance of its fully optimized behavior. Non-differentiable changes to the mechanical structure of each robot in the population -- mutations that rearrange, combine, add, or remove body parts -- were applied by a genetic algorithm in an outer loop of search, generating a continuous flow of novel morphologies with highly-coordinated and graceful behaviors honed by gradient descent. This enabled the exploration of several orders-of-magnitude more designs than all previous methods, despite the fact that robots here have the potential to be much more complex, in terms of number of independent motors, than those in prior studies. We found that evolution reliably produces ``increasingly differentiable'' robots: body plans that smooth the loss landscape in which learning operates and thereby provide better training paths toward performant behaviors. Finally, one of the highly differentiable morphologies discovered in simulation was realized as a physical robot and shown to retain its optimized behavior. This provides a cyberphysical platform to investigate the relationship between evolution and learning in biological systems and broadens our understanding of how a robot's physical structure can influence the ability to train policies for it. Videos and code at

A Study on Evolutionary Robotic System with Reconfigurable Morphology and Intelligence (mechanism of Emergent Symmetry Properties)

Environmental Adaptation of Robot Morphology and Control Through Real-World Evolution

Comprehensive Survey of Evolutionary Morphological Soft Robotic Systems

Scalable Co-Optimization of Morphology and Control in Embodied Machines

Morphological Symmetries in Robotics

Adjustbot: Bio-Inspired Quadruped Robot with Adjustable Posture and Undulated Body for Challenging Terradynamic Tasks

The Role of Morphological Variation in Evolutionary Robotics: Maximizing Performance and Robustness

Gait-learning with morphologically evolving robots generated by L-system

Emergence of behavior through morphology: a case study on an octopus inspired manipulator

Embodied Intelligence & Morphological Computation in Soft Robotics Community: Collaborations, Coordination, and Perspective

Automatic robot design inspired by evolution of vertebrates

Investigating Premature Convergence in Co-optimization of Morphology and Control in Evolved Virtual Soft Robots

On Designing a Learning Robot: Improving Morphology for Enhanced Task Performance and Learning

A Deep Reinforcement Learning Approach to Optimal Morphologies Generation in Reconfigurable Tiling Robots

Spontaneous Emergence of Multitasking in Minimal Robotic Systems.

An Investigation of the Factors Influencing Evolutionary Dynamics in the Joint Evolution of Robot Body and Control

Efficient and Diverse Generative Robot Designs using Evolution and Intrinsic Motivation

No-brainer: Morphological Computation driven Adaptive Behavior in Soft Robots

A Comparative Study of Brain Reproduction Methods for Morphologically Evolving Robots

Multi-environment robotic transitions through adaptive morphogenesis

Evolution and learning in differentiable robots