Differentiable Optimization-Based Modular Planning Framework for Pick-And-Place with Regrasp
Yejun Song, Seoki An, Somang Lee, Jeongmin Lee, Jeongseob Lee, Geun Su Yoo, Dongjun Lee
AI summary
Problem
Planning pick-and-place tasks for low-dexterity grippers often requires regrasp, but existing methods rely on inefficient sampling or computationally expensive physics simulations, especially in high-dimensional configuration spaces.
Approach
The framework decomposes planning into modular, differentiable optimization steps that jointly consider pick and place constraints, predict stable object poses via geometry optimization, and compute collision-free paths without sampling or dynamic simulation.
Key results
- Joint grasp generation across pick and place scenes without sampling
- Simulation-free stable pose prediction using differentiable support functions
- Non-static release planning via optimized basins of attraction
- Successful pick-and-place with repeated regrasp validated in simulation and hardware
Why it matters
Provides a scalable, efficient planning solution for industrial and construction robots that must manipulate objects with limited gripper dexterity.
Abstract
Robotic manipulation commonly involves pick- and-place tasks in which regrasp may be necessary for low- dexterity manipulators. Many existing approaches rely on sampling, which becomes inefficient when repeated regrasp is required in high-dimensional configuration spaces. We propose a modular planning framework that comprises differentiable optimization-based modules: grasp generation, stable pose pre- diction, inverse kinematics solving, and path planning. The modular design yields a systematic pipeline, enabling direct pick-and-place, static or non-static release, and repeated re- grasp by solving each module as needed. Each module leverages differentiable geometric features to efficiently solve its corre- sponding optimization problem. Our framework explicitly ac- counts for grasp constraints across both task scenes and predicts stable poses for regrasp planning via optimization rather than expensive physics simulations, thereby improving the feasibility and efficiency of planning. We validated the framework in pick- and-place simulations and real-world experiments.