MPGNet: Learning Move-Push-Grasping Synergy for Target-Oriented Grasping in Occluded Scenes

TL;DR

This paper introduces MPGNet, a three-branch neural network that learns to synergize moving, pushing, and grasping actions for efficient target-oriented grasping in occluded scenes, validated through simulations and real-world tests.

Contribution

The paper presents MPGNet, a novel multi-branch network with a multi-stage training strategy for integrated move-push-grasping in occluded environments, advancing target-oriented robotic manipulation.

Findings

Effective in occluded scenes with minimal manipulations

Outperforms existing methods in simulated and real-world tests

Demonstrates robust synergy between actions

Abstract

This paper focuses on target-oriented grasping in occluded scenes, where the target object is specified by a binary mask and the goal is to grasp the target object with as few robotic manipulations as possible. Most existing methods rely on a push-grasping synergy to complete this task. To deliver a more powerful target-oriented grasping pipeline, we present MPGNet, a three-branch network for learning a synergy between moving, pushing, and grasping actions. We also propose a multi-stage training strategy to train the MPGNet which contains three policy networks corresponding to the three actions. The effectiveness of our method is demonstrated via both simulated and real-world experiments.