This WASP Advanced Autonomous Systems course will extend the existing foundational course, Autonomous systems, with more advanced concepts such as MPC-based path tracking, frontier-based exploration of unknown environments, learning-based approaches for action and perception, high-level planning, high-level decision-making, robot-behaviour modeling with Robot Skills, Finite State Machine, Behaviour Trees and Planning Domain Description Language (PDDL) abstractions.
Course type:
Advanced course for PhD students
- AI track: elective
- AS track: elective
- Joint curriculum: advanced
Time: Given odd years, Fall
Teachers: George Nikolakopoulos (LTU), Olov Andersson (KTH), Volker Krueger (LU)
The course requires
- A background in the basics of Autonomous Systems as covered by the WASP-course “Autonomous Systems”.
- A background in AI and deep learning corresponding to as a minimum the contents of the WASP-course “AI and Machine Learning”.
- A solid programming experience in a high-level language; the programming assignments will use Python and ROS2.
On completion of the course, the student should be able to
- Understand fundamental low-level control architectures and their tuning.
- Understand how to design a Model Predictive Controller (MPC) for path tracking.
- Design and experimentally verify an MPC framework on a mobile robot.
- Understand the meaning, the creation and the selection of frontiers in autonomous exploration missions.
- Understand how to use and evaluate common learning-based methods for autonomous robots and vehicles.
- Explain the opportunities and challenges of different learning-based architectures as well as recent trends in large pre-trained models and embodied AI.
- Use hierarchical Finite State Machines and Behaviour Trees for task modeling.
- Use PDDL and Planning for high-level planning.
The WASP Autonomous Systems course covers the Sense-Reason-Action cycle of autonomous systems to some extent, incl. path planning, sensing and basic control.
This WASP Advanced Autonomous Systems course will extend the previous course with more advanced concepts such as MPC-based path tracking, frontier-based exploration of unknown environments, learning-based approaches for action and perception, high-level planning, high-level decision-making, robot-behaviour modeling with Robot Skills, Finite State Machine, Behaviour Trees and Planning Domain Description Language (PDDL) abstractions. The course consists of three modules hosted at LTU, KTH and LU, respectively.
The examination in the course consists of assignments to be completed individually and in groups. Some of the assignments or parts thereof will be carried out or examined during the in-person 2-day meetings to allow for more interaction and better support.
A re-examination will be made available, upon request, about 6 months after the course
covering completion of missing parts.