Walker 6-DoF Grasping

This repository contains the implementation of 6-DoF General Object Grasping for WalkerII Humanoid Robot with Multi-fingered hands. The overview of our grasping framework is shown below:

The relevant publication is listed as follows:

• Zhuo Li, Shiqi Li, Ke Han, Xiao Li, Youjun Xiong, and Zheng Xie. Planning Multi-fingered Grasps with Reachability Awareness in Unrestricted Workspace. Journal of Intelligent & Robotic Systems, 2022. [PDF ] [Video]

The next sections provide instructions for getting started with our work.

• Walker 6-DoF Grasping
  • Installation
  • Data Generation
    • Raw Data Generation
    • Raw Data Clean
    • Dataset Construction
  • Network Training
  • Simulated Grasping
  • Real-world Robotic Grasping
    • Dependencies
    • Start the Robot
    • Hand-eye Calibration
    • Grasp Prediction
    • Grasp Execution

Installation

The following instructions were tested with python3.8 on Ubuntu 20.04. A ROS installation is only required for visualizations and interfacing hardware. Simulations and network training should work just fine without. The Real-world Robotic Grasping section describes the setup for real-robot grasping experiments in more details.

OpenMPI is optionally used to distribute the data generation over multiple cores/machines.

sudo apt install libopenmpi-dev

Clone the repository into the src folder of a catkin workspace.

git clone https://github.com/RIP4KOBE/walker_6dof_grasping.git

Create and activate a new virtual environment.

cd /path/to/walker_6dof_grasping
python3 -m venv --system-site-packages .venv
source .venv/bin/activate

Install the Python dependencies within the activated virtual environment.

python3 -m pip install -r requirements.txt

Build and source the catkin workspace,

catkin build walker_6dof_grasping
source /path/to/catkin_ws/devel/setup.bash

gedit ~/.bashrc
export PYTHONPATH="/path/to/walker_6dof_grasping/src"

Data Generation

Raw Data Generation

Generate raw synthetic grasping trials using the pybullet physics simulator.

python scripts/generate_data.py data/raw/foo --sim-gui

• python scripts/generate_data.py -h prints a list with all the options.
• mpirun -np <num-workers> python ... will run multiple simulations in parallel.

The script will create the following file structure within data/raw/foo:

• grasps.csv contains the configuration, label, and associated scene for each grasp,
• scenes/<scene_id>.npz contains the synthetic sensor data of each scene.

Raw Data Clean

python scripts/raw_grasp_data_clean.py

The script is useful to clean and balance the generated grasp data.

Dataset Construction

python scripts/construct_dataset.py data/raw/foo data/datasets/foo

• The script will generate the voxel grids/grasp targets required to train GPN.
• Samples of the dataset can be visualized with the vis_sample.py script and gpn.rviz configuration. The script includes the option to apply a random affine transform to the input/target pair to check the data augmentation procedure.

Network Training

python scripts/train_gpn.py --dataset data/datasets/foo [--augment]

Training and validation metrics are logged to TensorBoard and can be accessed with

tensorboard --logdir data/runs

Simulated Grasping

Run simulated clutter removal experiments.

python scripts/sim_grasp.py --model data/models/walker2_model/gpn_conv_1545.pt --sim-gui --rviz

• python scripts/sim_grasp.py -h prints a complete list of optional arguments.
• To visualize the predicted grasps, you need to run the following code:

cd /path/to/walker_6dof_grasping/config/
rviz -d sim.rviz

• Use the clutter_removal.ipynb notebook to compute metrics and visualize failure cases of an experiment.

Real-world Robotic Grasping

This section contains the implementation of our method on physical WalkerII humanoid service robot.

Dependencies

easy_handeye

aruco_ros

realsense_ros

walker_kinematics_solver

Start the Robot

First, you need to start the WalkerII robot with the following conmands:

• launch walker_control

ssh [email protected]
aa
sudo -s
roslaunch walker_control walker_control.launch 

• launch whole_body_control

ssh [email protected]
aa
sudo -s
aa
cd ~/run
./autorun.sh

• launch service

rosservice call /walker/controller/enable "data: true"

• launch legs

ssh [email protected]
aa
sudo -s
aa
roslaunch leg_motion walker2_leg.launch

Hand-eye Calibration

Next, Using easy_handeye package to perform the hand-eye calibration (Note that you can use other hand-eye calibration methods, as long as you get the correct transformation relationship between the robot hand coordinate and the camera coordinate):

• Attach the Aruco Marker to the palm of the robot hand.
• Launch the camera

cd path/to/realsense_ros
roslaunch realsense2_camera rs2_camera.launch

• Launch easy_handeye

cd path/to/easy_handeye
roslaunch easy_handeye walker_aruco_calib_realsense.launch

• Sampling around 20 group of poses and recording the trasnformation matrix

Grasp Prediction

• Launch the camera

cd path/to/realsense_ros
roslaunch realsense2_camera rs2_camera.launch

• Predicting Feasible Grasps

cd path/to/walker_6dof_grasping
source .venv/bin/activate
python scripts/walker_detection_single_cam.py --model data/models/walker2_model/vgn_conv_1545.pt 

• Visualizing Prediction Results

cd path/to/walker_6dof_grasping/config
rviz -d sim.rviz 

Grasp Execution

• IK Solving and Reachable Grasps Publishing

cd path/to/walker_kinematics_solver
rosrun grasp_pose_ik_solver

• Grasp Execution

cd path/to/walker_kinematics_solver
rosrun walker_grasp

Remember to set the motion_segment flag as true when you first run the walker_grasp.cpp

line 548 bool motion_segment0 = true;