Language as a Sensor: CoRL 2026 Paper Fuses Speech With Vision for Robot Mapping
andreea7b · x · 2026-10-23
- Researchers present Language as a Sensor (CoRL 2026): robots treat human language like "I left my bag on the table" as a calibrated sensor input rather than relying only on onboard sensors for mapping.
- The method splits ambiguity in two: an LLM proposes hypotheses about what the speaker meant (e.g., which table), and a learned model predicts the target object's location for each hypothesis with calibrated uncertainty.
- These language-derived spatial beliefs are fused with vision, sharpening as the robot explores. Paper and project page are available.
More from Embodied
- After watching a robot practice parking, blogger wants Claude judging via overhead camera — _Stocko_ · 2026-10-23
- Dev hooks OpenAI Realtime API to a DIY robot arm: sim-trained skills plus vision for tic-tac-toe — Tarun122 · 2026-10-23
- Microduck visits ROBOTIS HQ to stock up on motors for more robot friends — TinfoilTricorn · 2026-10-23
- EPFL's LogSync neural chip decodes network-level signals between brain regions — MacrinePhD · 2026-10-23
- Wayve, Stellantis and NVIDIA partner with French government to build full-stack AV ecosystem — alexgkendall · 2026-10-23
- AI Is Making Hardware Learning Easier: Interactive 3D Learn Arduino App Gets Major Update — burhop · 2026-10-23