ROMA: RUC, BAAI and AresoX unveil multi-sensory active perception system for robots
jiqizhixin · x · 2026-10-17
Researchers from Renmin University of China (GeWu-Lab), BAAI, and AresoX present ROMA (Real-World Object-Centric Multi-Sensory Active Perception), an LLM system giving multi-sensor robots the ability to actively gather missing information in the physical world.
The motivating scenario: asked to "find me an empty cup for tea," a robot must infer implicit intent, then determine which candidate cup is actually empty — shaking one, weighing another, listening for liquid. ROMA addresses the core embodied gap of knowing what information is missing and which sensory modality to use to acquire it.
The release includes a complete stack: dataset, benchmark, and a multimodal reasoning model.
More from Embodied
- David Hoang plans to resurrect an HTC One as agentic hardware — davidhoang · 2026-10-17
- Maker 3D-prints robot joints inspired by a construction girl robot — _Stocko_ · 2026-10-17
- Erwin Coumans revives 2010-era CUDA physics engine with Claude in one afternoon — erwincoumans · 2026-10-17
- Teknium catalogs 330 hackable devices and 1,697 project ideas for Hermes Agent — Teknium · 2026-10-17
- Scobleizer visits TheHumanoidHub: robots for devs this year, everyone else after 2030 — Scobleizer · 2026-10-17
- NeodraDynamics unveils Neo1, a 4-meter 1.2-ton heavy-duty humanoid rated for 300 kg per arm — CyberRobooo · 2026-10-17