Humanoid robots fold, scoop and handle fragile objects more precisely with new tech

Carnegie Mellon University and Bosch Center for AI introduced HTD, a humanoid robot framework built for touch-aware manipulation.

HTD combines imitation learning, predictive tactile modeling, and split upper-body/lower-body control to better balance motion and contact.

In tests, the system improved success rates on tasks like folding cloth, scooping items, inserting objects, and serving tea versus the ACT baseline.

The approach could make humanoid robots more reliable for fragile, contact-heavy work in homes and workplaces.