PINT: Shared Predictive Intent Tokens Unify Reasoning, World Modeling, and Robot Actions
Xinyu Guo · hf · 2026-10-16
Researchers from the Institute of Automation, CAS, with Fudan and Tsinghua, propose PINT, a unified model for robot manipulation that combines vision-language reasoning, video world modeling, and action generation—addressing VLA's lack of future dynamics modeling and the split between world models and action experts.
The core is a learnable predictive intent token: in a two-stage training scheme, stage one freezes the VLM and action expert while training the intent token to predict future states, so multiple experts share one intent. Without embodied pretraining, PINT scores 79.5 OOD on LIBERO-Plus (vs. Fast-WAM 51.5, StarVLA 75.0) and 98.0 on LIBERO, also leading on RoboTwin 2.0 transfer. Code is open-sourced at ZGCA-HMI-Lab/PINT.
More from Embodied
- Agility's Digit 5 pairs Nyobolt fast-charging battery: 90 min work, 9 min charge — xmercury_one · 2026-10-16
- VLA-Δ: Zhejiang University Adds Physical State Transition Supervision to VLA Models — OmniAI-ZJU · 2026-10-16
- TouchScale releases 500 hours of unified egocentric vision-tactile data for scaling study — jiqizhixin · 2026-10-16
- Ghost raises $11M for Core, a personal AI computer that runs locally in your home — nick_linck · 2026-10-16
- Robot shows up at Costco to fetch the famous $1.50 hotdog — majidmanzarpour · 2026-10-16
- 15 Leju Kuavo humanoid robots run 2 months at Hangzhou auto-parts line, 98% success rate — CyberRobooo · 2026-10-16