Transferring Experience from Simulation to the Real World for Precise Pick-And-Place Tasks in Highly Cluttered Scenes

TL;DR

This paper presents a learning-based system that uses simulation-trained neural networks to accurately grasp and place multiple objects in cluttered scenes, successfully transferring skills from simulation to real-world scenarios.

Contribution

The introduction of PQ-Net, a neural network that estimates grasp quality and object pose in real-time, enabling precise pick-and-place in cluttered environments with domain transfer from simulation.

Findings

PQ-Net achieves 92 fps in pose and quality estimation.

The system successfully transfers from simulation to real-world tasks.

Outperforms other model-free approaches in grasp success rate.

Abstract

In this paper, we introduce a novel learning-based approach for grasping known rigid objects in highly cluttered scenes and precisely placing them based on depth images. Our Placement Quality Network (PQ-Net) estimates the object pose and the quality for each automatically generated grasp pose for multiple objects simultaneously at 92 fps in a single forward pass of a neural network. All grasping and placement trials are executed in a physics simulation and the gained experience is transferred to the real world using domain randomization. We demonstrate that our policy successfully transfers to the real world. PQ-Net outperforms other model-free approaches in terms of grasping success rate and automatically scales to new objects of arbitrary symmetry without any human intervention.