Discrete robot policies match or beat continuous action heads, challenging common belief
siddkaramcheti · x · 2026-10-22
New research challenges the belief that robot policies need continuous action heads: policies with the simplest discrete representations match or beat continuous ones across sim and real, single-step and diffusion/flow settings. Co-author maxsimchowitz frames it as a broader lesson on modality representation—discrete parametrizations learn geometry from data without being constrained by it, "exactly what the bitter lesson tells us to look for."
More from Embodied
- Why a 2020 Paper on Rotation Representations Is Called the Biggest Kinematics Breakthrough in a Decade — andrew_n_carr · 2026-10-22
- SuperMimic trains humanoid control on 129k motions with 1 GPU in 3 hours, cutting GPU-hours 1000x — kevin_zakka · 2026-10-22
- USTC demos SpiRobs, an octopus-inspired logarithmic spiral robot — burny_tech · 2026-10-22
- Hugging Face's Microducks robot tops 15,000 units sold, closes out Nvidia's European tour — huggingface · 2026-10-22
- FunCAD generates working, 3D-printable CAD mechanisms from a single text prompt — jonLorraine9 · 2026-10-22
- WSJ's Joanna Stern built a folding robot that fold-tested iPhone Duo 32,000 times overnight — SumitGup · 2026-10-22