UT Dallas Robotics Lab Pivots From Perception to Generalizable Manipulation Learning
YuXiang_IRVL · x · 2026-10-04
The head of the Intelligent Robotics and Vision Lab at UT Dallas announced an updated research direction: after years focused on perception, the lab will emphasize robot learning for manipulation, aiming to build more capable and generalizable robots.
The revamped research page highlights robot foundation models and vision-language-action (VLA) models, cross-embodiment learning, and unified representations of physical actions, plus reproducible real-world benchmarks. Recent work includes cross-embodiment manipulation via a unified hand action space (RSS 2026 Workshop), NAC neural action codecs for VLA models (CoRL 2026), the low-cost VLA-REPLICA benchmark (NeurIPS 2026), and the ROBOMETER robotic reward model.
More from Embodied
- Developer shows off smart clothing designed to capture your every movement — Scobleizer · 2026-10-04
- Shift Robotics Launches Moonwalkers Dusk Power Shoes That Learn Your Gait in 10 Steps for $1,199 — Scobleizer · 2026-10-04
- Anthropic study: robots cost-competitive for just 0.3% of US work hours — VraserX · 2026-10-04
- Steam Frame hands-on review: plenty to like, a few annoyances — smtabatabaie · 2026-10-04
- Early user says OpenAI's dot device replaced ChatGPT and Codex in daily workflow — shaunralston · 2026-10-04
- New Chinese embodied AI model tops global ranking with 91.9 on Meta-World — yogthos · 2026-10-04