EAPN: execution-aligned noise fixes mode switching in asynchronous replanning, 96.7% bimanual success
Di Wu · hf · 2026-10-07
Execution-Aligned Progressive Noise (EAPN) tackles mode switching and inconsistent continuation across action chunks in real-time generative robot policies caused by independent stochastic initialization during continuous asynchronous replanning.
- It propagates a shared noise trajectory across replanning steps, aligns it with actual execution displacement, and models temporal correlation within each chunk.
- Aligned stochastic history plus committed action context conditions subsequent generation, so new chunks continue from execution-consistent states instead of restarting from fresh noise.
- Results: improved multimodal consistency on D3IL, 88.59% average on Kinetix, robust LIBERO performance even under long inference delays; real-robot success of 90.0% on Object Storage and 96.7% on bimanual Cloth Folding.
More from Embodied
- 2023's ACT + ALOHA was the original physical AI era, says dev — lukas_m_ziegler · 2026-10-07
- Two-step flow denoising cuts VLA inference from 61.6ms to 22ms for real-time robot control — Di Wu · 2026-10-07
- CTP: single-pass multimodal robot policy hits 97.25% on LIBERO, cuts inference latency to 75.8ms — Di Wu · 2026-10-07
- Magic-W0: a structured world-action foundation model topping RoboDojo-Sim at 27.10 — Xuhua Chen · 2026-10-07
- RaiSim is back: v2.8.0 brings ultra-fast accurate CPU simulation plus photorealistic rendering — ChongZzZhang · 2026-10-07
- LiteReality-Agent turns images into simulation-ready 3D scenes for MuJoCo robot planning — elliottszwu · 2026-10-07