Aditya Patankar

Aditya Patankar

Postdoctoral Associate. Task specification, grasping and motion planning for robot manipulation.

Grasps and regrasps for multi-step tasks

Deciding whether one grasp can carry out a whole manipulation plan, and where the robot needs to regrasp when it can't.

IEEE International Conference on Robotics and Automation (ICRA), 2025. A. Patankar, D. Mahalingam, and N. Chakraborty.

PDF arXiv DOI Video

The problem

Many tasks are more than one motion. Pivoting a box several times in a row is a sequence of constant screw motions, and a grasp that works for the first motion may not work for a later one. The robot needs to know whether a single grasp will do for the whole plan, and if not, where to regrasp and which grasps to use.

Approach

We formalize regrasping in terms of the task. Starting from the object’s point cloud and a manipulation plan given as a sequence of constant screw motions, we use the task-dependent grasp metric to find which grasps can impart each motion. From this we compute whether a single grasp is enough for the whole plan or the robot needs to regrasp along the way, and synthesize the grasps.

A plan of three pivots with a regrasp before the last one
A plan with three pivots of a box: the robot grasps the box, pivots it twice, regrasps, and pivots it again.

Experiments

We ran the approach on a Franka Emika Panda. The clip at the top shows two trials, sped up 8×.

Papers