Reproduce “An Introduction to Zero-Order Optimization Techniques for Robotics” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Armand Jordana, Jianghan Zhang, Joseph Amigo, Ludovic Righetti

The authors present a mathematical framework viewing trajectory optimization and reinforcement learning methods as random search that uses sampled function evaluations to approximate gradient updates. They interpret MPPI through log-sum-exp smoothing, derive block-diagonal CMA updates, and extend smoothing to actor updates. Block-diagonal CMA matches full-covariance CMA’s cost performance on tested tasks; RL gains over TD3 are unclear.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.