[ros-users] Code release: Active Volumetric Reconstruction for Mobile Robots (mobile manipulators, flying, ground robots, etc.)

Davide Scaramuzza scaramuzza.davide at gmail.com
Wed May 25 14:52:28 UTC 2016


We are pleased to announce the release of a new open-source library for 
performing active object reconstruction using a camera-based depth 
sensor on a mobile robot. The rpg_ig_active_reconstruction framework is 
available from our github:

The key features are:
- General framework for reconstruction that is object, sensor, and 
- Real-time reconstruction of unknown objects that are spatially 
bounded: the user only needs to know the size of the bounding volume, 
and our algorithm will efficiently reconstruct its structure volumetrically.
- Modular library can be adapted to the kinematics of arbitrary mobile 
robots (ground, flying, mobile manipulators, etc.), and can receive 3D 
input from any dense depth sensor (e.g. stereo cameras, RGB-D sensors, 
monocular dense depth estimation [such as REMODE], etc.).
- Multiple information gain formulations are implemented to guide the 
choice of next best view.
- Volumetric representation allows estimation of information gain for 
candidate views using both occupancy and occlusion.
- Includes ROS bindings and a simulated example in Gazebo.
- Video showing the system in action: https://youtu.be/ZcJcsoGGqbA

Technical details of the algorithm and information gain formulations can 
be found in our ICRA paper:

Stefan Isler, Reza Sabzevari, Jeffrey Delmerico, Davide Scaramuzza

University of Zurich, Robotics and Perception Group

More information about the ros-users mailing list