Hierarchical Reactive Grasping Via Task-Space Velocity Fields and Joint-Space Quadratic Programming
Yonghyeon Lee, Tzu-Yuan Lin, Alexander Alexiev, Sangbae Kim
AI summary
Problem
Real-time, collision-free reactive grasping for high-dimensional arm-hand systems is computationally intractable due to the combinatorial explosion of state and planning dimensions, while existing methods often suffer from local minima, poor obstacle handling, or excessive computational cost.
Approach
The method decouples global guidance from local feasibility by computing low-dimensional task-space velocity fields from lightweight 3D path optimizations, then tracking them in full joint space via a weighted quadratic program that enforces all collision and kinematic constraints.
Key results
- Real-time collision-free reaching around concave objects and obstacles
- Robust reactive grasping under dynamic disturbances
- Successful validation on a 15-DoF arm-hand system in simulation and reality
- Avoidance of local minima via hierarchical task-space to joint-space decoupling
Why it matters
It enables reliable, real-time dexterous manipulation for complex robotic systems in unstructured environments, advancing the practical deployment of agile robots in dynamic industrial and service settings.
Abstract
We present a fast and reactive grasping frame- work that combines task-space velocity fields with joint-space Quadratic Program (QP) in a hierarchical structure. Reactive, collision-free global motion planning is particularly challenging for high-DoF systems, as simultaneous increases in state dimen- sionality and planning horizon trigger a combinatorial explo- sion of the search space, making real-time planning intractable. To address this, we plan globally in a lower-dimensional task space – such as fingertip positions – and track locally in the full joint space while enforcing all constraints. This approach is realized by constructing velocity fields in multiple task-space coordinates (or, in some cases, a subset of joint coordinates) and solving a weighted joint-space QP to compute joint velocities that track these fields with appropriately assigned priorities. Through simulation experiments and real-world tests using the recent pose-tracking algorithm FoundationPose [1], we verify that our method enables high-DoF arm–hand systems to perform real-time, collision-free reaching motions while adapting to dynamic environments and external disturbances. Project page: https://reactivegrasp.github.io.