Accelerating Visual-Policy Learning through Parallel Differentiable Simulation

TL;DR

This paper introduces a computationally efficient visual policy learning algorithm that uses differentiable simulation with decoupled rendering, resulting in faster training and improved performance on complex control tasks.

Contribution

The authors propose a novel decoupling approach for rendering in differentiable simulation, reducing computational overhead and enhancing policy gradient stability.

Findings

Significantly reduces training time on visual control benchmarks.

Outperforms baseline methods in final return metrics.

Achieves 4x improvement on humanoid locomotion tasks.

Abstract

In this work, we propose a computationally efficient algorithm for visual policy learning that leverages differentiable simulation and first-order analytical policy gradients. Our approach decouple the rendering process from the computation graph, enabling seamless integration with existing differentiable simulation ecosystems without the need for specialized differentiable rendering software. This decoupling not only reduces computational and memory overhead but also effectively attenuates the policy gradient norm, leading to more stable and smoother optimization. We evaluate our method on standard visual control benchmarks using modern GPU-accelerated simulation. Experiments show that our approach significantly reduces wall-clock training time and consistently outperforms all baseline methods in terms of final returns. Notably, on complex tasks such as humanoid locomotion, our method achieves a $4\times$ improvement in final return, and successfully learns a humanoid running policy within 4 hours on a single GPU.