What happened
Panasonic HD and Panasonic R&D Company of America, with UC Berkeley, NVIDIA, Stanford, Sapienza University of Rome and ItalAI researchers, built T-Rex, which combines a VLA model with fingertip touch data. Across 12 delicate two-handed tasks it averaged a 65% success rate — 30 points above existing methods.
Why it matters
T-Rex corrects its motions in real time as contact conditions change, enabling delicate force control and handling of soft objects — capabilities that could support manufacturing and logistics work where fine force adjustment is needed.
What to watch
The team will present the research at the Conference on Robot Learning (CoRL) in Austin from November 9 to 12.
WHO IT HITSManufacturing and logistics operators stand to benefit, as T-Rex is aimed at tasks like part insertion, screw tightening, and handling fragile items — work that requires fine force control and could ease worker burden and labor shortages.
Summaries like this, every morning.
T-Rex builds on the vision-language-action (VLA) approach, which generates robot actions from visual and language inputs. The team's key addition is fingertip tactile sensing, which lets the model adjust actions as contact changes. To train it, the researchers collected over 7,700 two-handed robot manipulation episodes totaling 100 hours, using more than 200 everyday objects and 22 basic actions, with synchronized camera, robot state, tactile, and language data. Looking ahead, Panasonic aims to apply the technology to manufacturing tasks like part insertion and screw tightening, and to logistics handling of flexible or fragile items, targeting reduced worker burden and alleviation of labor shortages.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
On October 21, 2026, Anfini CEO Jun Hashimoto gave a special lecture at Meiji University's engineering faculty…
In one session that ran 13 days and 5,065 requests, a single request late in the session cost about $0.42 agai…
A six-year web designer who signed up for Claude Pro in April 2026 and later moved to Max published the first…
EC自動化工房 laid out four things to standardize before AI translation, using a sample mug "350ml" that mixed unit…
At WSJ Tech on Nov. 4, OpenAI president Greg Brockman said annualized revenue grew 70% in Q3 2026, with enterp…
Supasaito said it confirmed on November 4, 2026 that its listing appears in ChatGPT's directory, describing co…
