Robotics researchers clash: RL controllers fail out of distribution, yet remain the pragmatic default
KyleMorgenstein · x · 2026-10-02
Robotics researcher KyleMorgenstein pushes back on claims about "unmodeled errors" in a debate on RL vs classical control: domain randomization is itself an explicit noise model, and RL controllers are famously poor out of distribution. Still, he concedes that since most people are bad at controls and system modeling, RL with copy-paste rewards and hyperparameters is strictly easier and the right tool most of the time — the original claims were just overstated.
Related event: Robotics Researchers Debate Whether RL Beats Classical Control(4 posts)→
More from Embodied
- Hugging Face LeRobot adds humanoid support: π0.5 policy drives 29-DoF Unitree G1 — freelerobot · 2026-10-02
- Researcher's blunt verdict on humanoids: "The robots do not work" — andrew_n_carr · 2026-10-02
- Robot's Early Taichi Policy Trained on Kaggle Shows Promising Kung Fu Moves — freelerobot · 2026-10-02
- BeyondSCe grasps objects by past-event reference, hitting 77% success on occluded targets — SeoulNationalUniv · 2026-10-02
- NUS's APPL uses structural priors to bridge skill learning and composition in robot manipulation — NationalUniversityofSingapore · 2026-10-02
- Open-source Harness fork moves coding agents out of the app into orca — dee_hw · 2026-10-02