Thesis defense: Vindula Jayawardana
45-600B
Doctoral Thesis Title: Learning to Tackle Task Variations in Control – A Transportation Context
Presenter: Vindula Jayawardana
Presenter’s Affiliation (CSAIL, RLE, LIDS, MTL, etc.): LIDS
Thesis Supervisor(s): Cathy Wu
Date: 12th June 2025
Time: 10 AM
Location if in person: 45-600B
Abstract: Real-world control tasks are messy and often exhibit task variations. Practical solutions to these problems must exhibit generalization across task variations. For example, in the task of controlling traffic signals, control strategies must adapt to different intersection topologies (the variations), each with distinct dynamics. In this thesis, we consider the challenge of coping with task variations in the context of transportation problems-specifically in roadway interventions- where many such variations are both common and imperative to handle. We develop machine learning techniques to address three key challenges: 1) quantify the impact of task variations in control, 2) model those variations to align with the real world, and 3) optimize in the presence of them. To this end, we first emphasize the importance of explicitly accounting for task variations in problem specifications. We highlight that task underspecification—resulting from loosely defined variations—can significantly hinder effective decision-making. Then, we propose IntersectionZoo, a benchmark designed to drive the development of learning algorithms that can generalize across task variations. We further explore the possibility of task variation modeling from a generative modeling point of view, using human driver behavior modeling as a case study. Next, we present a contextual reinforcement learning algorithm that can learn to explicitly leverage task variation structure and generalize effectively in an emerging application of cooperative eco-driving with autonomous vehicles. Finally, we translate these efforts into practical applications and illustrate how explicitly modeling task variations can surface otherwise overlooked insights, as shown in a metropolitan-scale cooperative eco-driving case study aimed at supporting climate change mitigation goals. More broadly, this thesis paves the way for large-scale roadway interventions that are robust, adaptable, and aligned with sustainability goals.
Details
- Date: Thursday, June 12
- Time: 10:00 am - 12:00 pm
- Category: Thesis Defense
- Location: 45-600B