CTP: single-pass multimodal robot policy hits 97.25% on LIBERO, cuts inference latency to 75.8ms
Di Wu · hf · 2026-10-07
Conditional Trajectory Peaks (CTP) is a single-pass robot policy framework that jointly predicts complete action-chunk candidates, probability masses, and trajectory scales, addressing multimodal imitation learning and cross-replanning consistency.
- DAPS specializes trajectory peaks via trajectory-level posterior responsibilities with mass- and scale-modulated overlap constraints.
- ETBT maintains cross-chunk consistency through geometric correspondence between exchangeable candidate sets, letting current policy evidence override historical constraints.
- Results: 91.40% coverage on Push-T; 100.0%/79.72%/84.44% on D3IL Avoiding/Aligning/Sorting-2; 97.25% average on LIBERO; 50/50 success on a real dual-arm two-plate task.
- Matches π0.5 on bottle uprighting and pen placement while cutting inference latency from 218.24ms to 75.80ms.
More from Embodied
- 2023's ACT + ALOHA was the original physical AI era, says dev — lukas_m_ziegler · 2026-10-07
- EAPN: execution-aligned noise fixes mode switching in asynchronous replanning, 96.7% bimanual success — Di Wu · 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
- 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