GraspTune improves robot grasp success by using touch feedback
GraspTune: Tactile-Driven Execution Refinement for Robust Grasping
Robotics
Summary
Robots can plan where to grab objects using cameras, but actually holding them securely is still hard. The authors created GraspTune, a system that adjusts a robot’s hand movements based on touch sensors during the grab to make the grasp more stable. Tested on many objects in simulations and real robots, GraspTune significantly increased the success rate of grasping, even with new objects the system had not seen before. This approach helps make robot grasps that look good visually into ones that work reliably in the real world.
What this means in practice
- •For robotics engineers: Improve the stability of robot grasps by integrating tactile feedback to refine hand motions during grasp execution.
- •For warehouse automation teams: Increase the success rate of robotic picking systems handling diverse and unseen objects by enhancing grasp robustness with tactile-driven refinement.
Authors
Juntao Li, Xingke Xia, Sichao Liu, Daqiang Guo
Abstract
Visual grasp proposal generation has advanced rapidly, yet converting a selected proposal into a stable physical grasp remains a central execution-stage challenge. This paper introduces GraspTune, a tactile-driven execution-stage refinement framework that starts from a nominal proposal and applies bounded residual TCP motions during approach, contact formation, and final grasp execution. GraspTune learns control-facing contact semantics from local depth, tactile signals, state, and history using state-conditioned expert contact queries and multi-task supervision for contact change, contact risk, and post-close readiness. The representation conditions a diffusion-pretrained residual policy and is aligned with PPO for closed-loop execution. Across more than 60,000 simulated executions over 20 object categories, GraspTune establishes an execution-layer benefit across four proposal generators, raising stable grasp success by +19.22, +9.55, +12.45, and +20.70 percentage points for GraspNet, Contact-GraspNet, AnyGrasp, and VGN. A four-fold held-out category study raises unseen-object execution from 54.58% to 70.33%, showing category-disjoint generalization of contact correction. Across more than 1,000 real-robot trials on a UR5e setup with Xense fingertip sensors, GraspTune raises GraspNet execution from 71.0% to 84.3%, validating direct transfer without realworld policy fine-tuning. Together, these results turn visually plausible proposals into stable physical grasps for downstream contact-rich manipulation. A supplementary video is available at https://youtu.be/kcq7fSLNtzU.