And now, what's going to happen to us without barbarians?
They were, those people, a kind of solution.
[C. Cavafy, "Waiting for the Barbarians"]
Openings
- One PhD Studenship in Humanoid Perception
Official Link: IIT-PhD Opening Call and PhD_Calls.pdf
Where: Department of Advanced Robotics (ADVR-IIT) at the Istituto Italiano di Tecnologia (IIT), in Genoa, Italy.
Starting Date: November 2014.
The deadline for applications is: August 30th, 2014 noon (Italian time/CET).
Title: 17. 3D Perception for Humanoid Robots: Rough terrain locomotion and Free-
form object manipulation.
Tutors: Dr. Dimitrios Kanoulas, Dr. Nikos Tsagarakis
Description: After Fukushima Daiichi nuclear disaster in 2011 the need of robots to deal with unstructured environments and replace humans in hazardous tasks became one of the main open problems in robotics. Rapid advancements in actuation and control over the last few years enabled articulated humanoid robots to both walk in uneven terrain and perform dexterous manipulation using their hands. These capabilities are usually gained without using 3D perception at all, by assuming that either the environment is mostly known and well-structured, or the uncertainty can be tolerated by low-level feedback control. In real world scenarios these assumptions may not hold. We need vision! The problem of foot placement in
rough terrain (for example in a rocky trail) for walking or the problem of grasping free-formed objects (for example a rock) using 3D perception remains one of the central challenges in robotics and is the key aspect for completing locomotion or manipulation tasks in unknown environments.
The aim of this topic is to develop new environment reconstruction techniques that enable humanoid robots to perform both legged locomotion and manipulation tasks in unstructured environments using 3D perception for foot or hand placement. The state-of-the-art 3D perception sensors will be used (stereo/time-of-flight cameras, laser sensors, or structured light systems) along with other perception sensors like tactile, force control, or IMU ones. The dense 3D point cloud that is acquired from a range sensor will require some geometric simplifications for reasoning the contact between the robot's foot/hand and an area in the environment. Modeling these contact areas around and on a robot while using Simultaneous Localization and Mapping (SLAM) techniques for creating and keeping a map of these with respect to
the robot is a key aspect for completing these tasks. The developed methods will be tested both in simulation and on a real humanoid robot (COMAN). The project is interdisciplinary since perception needs to be combined with path planning and control techniques for making the actual robot complete a task. Thus the collaboration with other members of the project will be required into that direction. The work activity of this theme will be in line with the developments of the WALK-MAN EU project (http://www.walk-man.eu/).
Requirements: This topic lies in the intersection of Computer Vision and Robotics. Ideal applicants should have strong analytical and programming skills (C/C++ and MATLAB). A relevant degree is required, for instance in Computer Science or Engineering. A background in Robotics/Computer Vision is desirable, while knowledge of the Robot Operating System (ROS), the Point Cloud Library (PCL), or the Open Source Computer Vision Library (OpenCV) is a big plus. The applicants should be fluent in English and team players.
Reference: Dimitrios Kanoulas and Marsette Vona, Bio-Inspired Rough Terrain Contact Patch Perception, In the IEEE International Conference on Robotics and Automation, ICRA 2014.
Contacts: nikos.tsagarakis@iit.it