RoboSpan researchers report higher success in coordinated robot tasks

Researchers behind the RoboSpan preprint report over 20% higher task success rate than mainstream baselines in experiments on a physical robot with 29 degrees of freedom. The tests covered extended tasks requiring coordinated movement of the mobile base, arm and hand. Their framework separates control into hand manipulation, arm positioning and mobile-base movement, while allowing those branches to exchange information. A vision-language model breaks tasks into stages to reduce accumulated errors, and a tactile representation captures contact signals across the palm. The reported improvement comes from physical robot experiments, with comparisons limited to the tested tasks and baselines.