Fast map learning with mobile robots

Mobile robots often need to learn an initially uknown map of a position-dependent parameter from sampled values. Examples include learning radio transmission rates, the density of litter at each point, forest density, etc. Moreover, learning this map is often only one part of the robot's task -- the robot may also have a navigation objective, low energy consumption goals, etc. In this project, we aim to design and study a robot motion control strategy that optimally takes into account both map learning and the other objectives of the robot. Encouraging real results (click for video) are already available for indoor radio map learning, with a Parrot AR.Drone 2 drone available in the ROCON lab. We have several objectives following from this:

• Novel algorithms based on global optimization, with convergence guarantees.
• Design and adapt algorithms for new use cases in litter detection. This is different from the radio map case because there the drone also had to transmit a data buffer, whereas here the primary objective is just to find a good map (later on, we may also attempt to collect some litter with another robot).
• Apply the algorithms outdoor with a DJI Matrice M200 or indoor with a Parrot Mambo (both available in the lab).
• Extend either the radio or litter-map algorithm to multiple agents.

We are looking for students with strong programming skills (with robotics and real-time systems experience a plus), and/or strong mathematical and analytical skills.

Apply by contacting Lucian Busoniu. This project is part of a cooperation with several universities in Europe, including e.g. TU Munich and the University of Dubrovnik, Croatia. Visits abroad are possible.