PhD position at the Max Planck Institute for Biological Cybernetics, Tübingen, Germany

The Max Planck Society (MPS) is Germany's most successful research organization. Since its establishment in 1948, it has produced 17 Nobel laureates from the ranks of its scientists, putting it on a par with the best and most prestigious research institutions worldwide. Max Planck Institutes are built up solely around the world's leading researchers. The Max Planck Institute for Biological Cybernetics (MPI-KYB) at Tübingen, Germany, is one of the 82 independently organized research facilities of the MPS that carry out basic research in the service of the general public in the fields of natural sciences, life sciences, social sciences, and humanities. The Department of Human Perception, Cognition and Action at MPI-KYB, directed by Prof. Dr. Heinrich H. Bülthoff, studies human perception with the help of virtual reality (VR). This enables the experiments to be conducted in a controlled and yet natural surroundings for which there are special hardware and experimental constructions. Among the several state-of-the-art laboratories at MPI-KYB is the Cyberneum, a VR research facility equipped with several sophisticated VR systems that provide unique opportunities to study human perception and human-machine interactions. The Autonomous Robotics and Human-Machine Systems (AR-HMS) group, within the department of Human Perception, Cognition and Action at MPI-KYB, opens a PhD position for motivated candidates with excellent qualifications.

Motivation

As robots increasingly become a part of daily life in our society, there is a preeminent need for the robots to be able to perceive humans in a more implicit manner in order to make human-robot interaction as natural as possible. For instance, in crowded urban environments, robots must use their own vision (or a network of sensors in the environment) to distinguish between those humans who require the robot's attention for some cooperative/collaborative purpose and those humans who simply occupy the same environment for other possible reasons. Such visual classification based on gesture recognition, emotion detection, etc., should precede any subsequent direct interaction (e.g., verbal) between the robot and a human in order to make the overall human-robot interaction natural and less complicated for the untrained human users of the robots. Simultaneously, diverse vision-based functionalities in robots are essential to accomplish complex tasks by human-robot or robot-only teams that involve interaction and/or collaboration. Such functionalities can range from simpler ones, e.g., single object or person detection, recognition and tracking using a single static camera to more complex ones, e.g., tracking multitude of people in crowded and highly dynamic environments and at the same time perceiving the emotional response of humans with whom the robot is directly interacting. Thanks to networked robot systems (NRS), presence of multiple mobile sensors (e.g., micro aerial vehicles equipped with camera) or static sensors (e.g., wall/ceiling mounted network cameras) provide a strong foundation to tackle such complex functionalities for real time applications. The focus of this thesis will, therefore, be on the issues of scalability and real time applicability of multiple vision-based functionalities in an NRS where human-robot interaction is one of the most essential components.

Keywords

Sensor fusion, Cooperative perception, Person tracking; detection and tracking from non-inertial frames; face and gesture recognition; stereo-vision systems; motion capture systems; human-robot interaction, multi-robot systems.

Summary of Global Objectives

Expected objectives of this PhD thesis are:

• To conceptualize and develop a framework that hierarchically integrates person detection, classification and tracking with face and gesture recognition, within an NRS that consists of human-sized mobile robots, micro aerial vehicles, static sensors in the environment and humans cooperating with the robot in an urban environmental setting. Primary focus will be on indoor scenarios.
• To conceptualize and develop novel algorithms for the vision-based functionalities embedded within the above mentioned framework. The major focus here will be on the scalability issues of those algorithms such that they are applicable to extremely large environments consisting of a high number of static and mobile sensors. Applicability refers to computational feasibility in real time while simultaneously maintaining optimality of the solution.

Expected Qualifications and Skills of the Candidate

• We seek highly qualified candidates with a master degree in one of the following broad areas: robotics, mechanical engineering, electrical engineering, computer science or other related fields.
• The candidate should have fluent command of English as a written and spoken language.
• Prior experience in computer vision and image processing is essential.
• The candidate must have excellent programming skills in one or more languages, e.g., C, C++ and python.
• Knowledge and experience in Robot Operating System (ROS) will be a plus.

Selection Procedure

Interested candidates who meet the above mentioned requirements should send the following documents as soon as possible (all in pdf format) to aahmad@isr.ist.utl.pt by the 31st of July 2014

• Motivation letter.
• Curriculum Vitae (Including a list of publications)
• Online link to their own code snippets or softwares developed (these can be inserted as a section in the CV).
• A 2-page summary of their master thesis or any other research results (which they consider as their most important results) (Bibliography should not be within these 2-page limit)
• Copy of the last diploma and transcripts (grade sheet).

Selected candidate will be expected to enroll in the PhD program in the beginning of September 2014 at the University of Tübingen and will carry out their research work at Max Planck Institute for Biological Cybernetics, Tübingen.

Other Information

Homepage of Max Planck Institute for Biological Cybernetics (MPI-KYB) at Tübingen, Germany. http://www.kyb.tuebingen.mpg.de/

Brief Description of Work

In the context of this PhD thesis work, a Network Robot System (NRS) will consist of i) a mobile robot with an omni-directional chassis equipped with vision sensors and simple actuators (arm/gripper), ii) multiple micro aerial vehicles (MAVs), and iii) static sensors fixed within the environment, e.g., network cameras.

Highly Scalable Sensor Fusion

To achieve robust vision-based functionalities through an NRS, one needs to perform optimal sensor fusion. However, as environments scale up in size and feature-richness, the amount of visual information that needs to be processed becomes overwhelmingly high. Consequently, performing sensor fusion optimally and in real-time becomes exponentially heavy. One good example is how the number of particles required by a particle filter-based (an approximately optimal technique) object tracker grow exponentially with the increase of the state space dimension to maintain a given accuracy of the tracker. Nevertheless, there are possible ways, e.g, exploiting dependencies between state variables, through which an increase in computational complexity can be restricted. In this PhD work, such techniques will be explored to develop highly scalable sensor fusion algorithms.

Implicit-and-Explicit Interaction

Another major focus of this work is to investigate methods for implicit human-robot interaction. Here, implicit interaction refers to embodied communication between humans and robots. Robots' understanding of human body/hand gestures, visual cues and human emotions based on facial expressions and body posture are among some forms of embodied communication that would eventually make human-robot interaction more fluid and natural. To this end, state-of-the-art vision-based techniques will be investigated for human body/hand gestures and emotion detection. Indeed, taking advantage of an NRS will facilitate the detection process, however, innovative algorithms must be developed for fusing visual information through various sensors available in the environment for this purpose.

On the other hand, explicit interaction between humans and robots involve activities such as voice-based communication, touch screen-based communication, etc. Humans naturally use both implicit and explicit form of communication in a general interaction. To this effect, fusion of visual information with that obtained through speech (microphones) and touch (touch screen) will be made. A hierarchical information fusion architecture will form the backbone of such human-robot interaction method.

Case Studies

Real robot implementation of the algorithms developed during this PhD thesis will be made in the following contexts: i) A domestic service robot (with an omni-directional mobile base) assisting an elderly person at home where the home environment will consist of static sensors as well as multiple MAVs with on-board sensors. ii) A service robot (same platform as in the first case study) assisting shoppers in a supermarket where the environment consists of several other robots of the same kind, multiple MAVs and static sensors.