Guiding Evolutionary Strategies by Differentiable Robot Simulators

TL;DR

This paper proposes combining Differentiable Robot Simulators with Evolutionary Strategies to significantly reduce sample complexity in robotic policy search, demonstrating 3-5 times efficiency gains in simulation and real-world experiments.

Contribution

It introduces a novel method that integrates DRS gradients with Evolutionary Strategies, improving sample efficiency in robotic policy optimization.

Findings

Sample complexity reduced by 3-5x in simulations.

Effective in both simulation and real-world environments.

Demonstrates potential for more efficient robotic learning.

Abstract

In recent years, Evolutionary Strategies were actively explored in robotic tasks for policy search as they provide a simpler alternative to reinforcement learning algorithms. However, this class of algorithms is often claimed to be extremely sample-inefficient. On the other hand, there is a growing interest in Differentiable Robot Simulators (DRS) as they potentially can find successful policies with only a handful of trajectories. But the resulting gradient is not always useful for the first-order optimization. In this work, we demonstrate how DRS gradient can be used in conjunction with Evolutionary Strategies. Preliminary results suggest that this combination can reduce sample complexity of Evolutionary Strategies by 3x-5x times in both simulation and the real world.