Dense Point-wise Line Voting for Robust 6D Pose Estimation in Industrial Bin-Picking

This repository contains authors' implementation of PLVNet: Dense Point-wise Line Voting for Robust 6D Pose Estimation in Industrial Bin-Picking. Our implementation is based on PointGroup and GPVPose. We warmly welcome any discussions related to our implementation and our paper. Please feel free to open an issue.

Introduction

Accurate 6D object pose estimation is crucial for robotic grasping and manipulation, particularly in industrial bin-picking scenarios. Despite challenges posed by heavy occlusion and object symmetries, we propose a novel approach based on point cloud inputs. Our method consists of two stages: first, a modified 3D-UNet is employed for instance segmentation, incorporating visibility prediction to mitigate occlusion effects. Subsequently, a point-wise line voting network is introduced to regress offset vectors towards object axis lines, aided by a RANSAC-based line fitting technique for robust pose estimation. Experiments on public datasets and real-world environments demonstrate the effectiveness of our approach in accurately estimating poses of objects in industrial bin-picking scenes, outperforming several baselines.

Environment setup

• Setup the python environtment

conda create -n plvnet python=3.7
conda activate plvnet
 
# install torch 1.8.1 built from cuda 11.1
pip install torch==1.8.1+cu111 torchvision==0.9.1+cu111 -f https://download.pytorch.org/whl/torch_stable.html

pip install numpy scipy opencv pyrender open3d pybullet

Running

1. Data generation and preprocessing

   python data_generate.py --dataset Sileance_Dataset --model_name gear --min_inst_num 1 --max_inst_num 60 --num_cycles_train 100 --num_cycles_val 10
   python data_preprocess.py --dataset Sileance_Dataset --model_name gear
   

2. Training

   # Config the file PointGroup/config/config_pointgroup.yaml then run the following command:
   python trainer_pointgroup.py
   # Config the file PoseEstimation/config/config_pose.yaml then run the following command:
   python trainer_gscpose.py
   

3. Evaluating

   python predict.py