Closed-Loop Next-Best-View Planning for Target-Driven Grasping

This repository contains the implementation of our IROS 2022 submission, "Closed-Loop Next-Best-View Planning for Target-Driven Grasping". [Paper][Video]

Setup

The experiments were conducted with a Franka Emika Panda arm and a Realsense D435 attached to the wrist of the robot. The code was developed and tested on Ubuntu 20.04 with ROS Noetic. It depends on the following external packages:

• MoveIt
• robot_helpers
• TRAC-IK
• VGN
• franka_ros and realsense2_camera (only required for hardware experiments)

Additional Python dependencies can be installed with

pip install -r requirements.txt

Run catkin build active_grasp to build the package.

Finally, download the assets folder and extract it inside the repository.

Experiments

Start a roscore.

roscore

To run simulation experiments.

roslaunch active_grasp env.launch sim:=true
python3 scripts/run.py nbv

To run real-world experiments.

roslaunch active_grasp hw.launch
roslaunch active_grasp env.launch sim:=false
python3 scripts/run.py nbv --wait-for-input