Pick and Place Planning is Better than Pick Planning then Place Planning

TL;DR

A modular algorithm for joint pick and place planning that can make use of state of the art grasp classifiers for planning multi-fingered grasps for novel objects from partial view point clouds is presented.

Abstract

Robotic pick and place stands at the heart of autonomous manipulation. When conducted in cluttered or complex environments robots must jointly reason about the selected grasp and desired placement locations to ensure success. While several works have examined this joint pick-and-place problem, none have fully leveraged recent learning-based approaches for multi-fingered grasp planning. We present a modular algorithm for joint pick and place planning that can make use of state of the art grasp classifiers for planning multi-fingered grasps for novel objects from partial view point clouds. We demonstrate our joint pick and place formulation with several costs associated with different placement tasks. Experiments on pick and place tasks with cluttered scenes using a physical robot show that our joint inference method is more successful than a sequential pick then place approach, while also achieving better placement configurations.

Figures (13)

• Figure 2: Factor graph of the pick and place probability distribution. We see that while the success probabilities are conditionally independent given the planning parameters, they can not fully decouple, requiring joint inference over pick and place parameters.
• Figure 3: Robot-object geometry for pick-and-place collision checking: (\ref{['fig:aug_obj']}) object; (\ref{['fig:aug_robot']}) robot; (\ref{['fig:aug_full']}) union of object and robot.
• Figure 4: Overview of our grasp prediction pipeline based on Grasp-Hermans-Lu-VoxelInf-RAM-2020. A grasp classifier predicts grasp success given an object voxel grid and grasp configuration. A mixture density network models a distribution over grasp configurations given an input voxel grid.
• Figure 5: Steps to generate signed-distance function for place scene and computing collision-free place initializations.
• Figure 6: Mean configurations of the MDN top and side grasp modes, visualized with the partial view point clouds of the objects (\ref{['fig:mdn_lego']}) lego blocks, (\ref{['fig:mdn_cracker']}) cracker box, (\ref{['fig:mdn_mustard']}) mustard bottle and (\ref{['fig:mdn_pitcher']}) pitcher.