New Project, New Challenges

After 6 years of working in the AI research group, I have taken up a research fellow postdoc position in the real-time and distributed systems group, specifically working on the SCHEME project (safety-critical harsh environment micro-processor evolution). I am applying many of the techniques from previous work such as feature engineering, algorithm selection, constraint programming in order to solve problems in robotics system.

The first paper on this new research area (for me) was accepted to the TAROS 2026 conference. The work addresses the issue of a robot carrying out a collection of tasks, with the possibility of using different motion planners for each sub-task; the focus is on predictability of performance and in particular the reliability of the combined solution in terms of meeeting deadlines.

Figure: Different algorithms perform better on different tasks

Planning algorithms show complementarity of performance across a range of tasks; by training a machine learning model to select an algorithm according to the features of a new task, we are able to show that the overall performance of the global task is more predictable, with a much shorter right tail in our evaluation simulations.