Many practical robotic problems (e.g., assembling the product, fetching the milk from the fridge, etc.) can be represented as sequential decision-making problems. Two major paradigms for solving sequential decision-making problems are planning and learning. Learning methods generally struggle in problems with long-term consequences of actions and we require deliberate long-horizon planning to solve such problems. On the other hand, planning in these (usually high-dimensional) search spaces is not trivial due to combinatorial explosion and the need for accurate models. To plan efficiently, it is often useful to use hierarchical abstractions. One way is to represent the problem as integrated Task and Motion Planning (TAMP) which consists of discrete modes (tasks) and continuous motions. Such hybrid problem representation allows for solving, on one hand, multimodal problems where local trajectory optimization approaches fail, and on the other hand, highly-constrained problems where decoupled “task planning followed by motion planning” is too conservative. In this talk, I focus on Task and Motion Planning problem representation and a search-based method (i.e. A* search) to solve them. This is demonstrated in several examples including automated driving in the urban environment, agile driving on slippery roads (i.e. drifting) and robot manipulation. Finally, we explore an effective combination of planning and learning methods together with some outlooks.

 

Bio:
Dr. Zlatan Ajanovic is a Postdoctoral Researcher in the Cognitive Robotics Department (CoR), Delft University of Technology (Netherlands) with Prof. Jens Kober. He received a master's degree from the Univesity of Sarajevo (Bosnia and Herzegovina) in 2015 and a PhD degree from the Graz University of Technology (Austria) in 2019, all focused on Automation and Control. Previously, he was a Marie Skłodowska-Curie Fellow and Senior Researcher at VIRTUAL VEHICLE Research GmbH, Graz (Austria), and a visiting researcher at TU Delft, University of Sarajevo, AVL List and Volvo Cars. He serves as a member of the IFAC Technical Committees for Optimal Control, Computational Intelligence in Control and Intelligent Autonomous Vehicles as well as the IEEE Technical Committee on Intelligent Control. He serves as a member of the organizing committee of the BeNeLearn/BNAIC 2023 and ICAPS 2024 as well as an Associate Editor for IROS 2023. His main research interest lies in the Decision-making and Control of Autonomous Robots based on Planning, Learning and Control Theory. He is the recipient of the IFAC Young Author Award and Hans List Award for his PhD thesis.
https://zlatanajanovic.com/