Master Thesis: Safe Planning with Optimal Control and Conformal Machine learning

Job description

Safe Planning for Autonomous Vehicles with Optimal Control and Conformal Machine Learning

 

Suggested Time Schedule: Jan 2025 – Jun 2025

 

More than one million people die in traffic accidents every year. One solution to improve traffic safety could be to develop autonomous vehicles (AVs) that avoid dangerous human driving behaviors [1]. Much progress has been made in recent years following the surge of machine learning driven control methods, showing great potential in predicting properties of very complex traffic situations. However, machine learning methods (ML) are infamous for lacking robustness when generalizing to unseen, out-of-sample data. Hence, many learning-based control applications lack the appropriate safety-guarantees such as, probabilistic feasibility. For safety-critical applications, such as autonomous heavy-vehicle-combinations (HVCs) where other road-users are extremely vulnerable, these safety-guarantees are of utmost importance.

An appealing approach to address safety is to plan a trajectory using Model Predictive Control (MPC). These approaches have gained attention due to the ability to enforce rigorous constraints on the ego-vehicle and collision avoidance [3,4]. More recent work has shown strong theoretical results that provide probabilistic feasibility guarantees for learning-based MPC, utilizing Conformal Prediction [5]. In this work, we aim to utilize the methods in [5] with collision avoidance constraints inspired by [6] to provide safe, learning-based trajectory planning for an autonomous HVC.

The purpose of this thesis can be summarized as:

1.    Study relevant literature in Safe Machine Learning, such as Conformal Prediction, and Collision Avoidance modelling.
2.    Formulate a stochastic optimal control problem that incorporates ML such that probabilistic guarantees are obtained for collision avoidance.
3.    Implement the Stochastic Model Predictive Controller with a suitable learning-based motion predictor (RNN, transformer, …) and evaluate it in a challenging highway-driving scenario.
4.    (Optional) A stretch goal for the project is to condense the thesis into a conference paper.

The aim of this thesis is to use stochastic MPC and machine learning to solve problems within autonomous driving. The work will include machine learning, stochastic optimization and programing. This work will be carried out together with Volvo Group Trucks Technology. The thesis is recommended for one or two students with a strong background in optimization, machine learning and Python/C++ with a solid mathematical background. Prior experience with vehicle modelling and simulation is meritorious. 

 

If you find this proposal interesting send your application (CV and grades) to:
leo.laine@volvo.com

Contact persons:
Leo Laine – Volvo GTT/ Chalmers University of Technology
tel: +46 31 323 53 11
mail: leo.laine@volvo.com

Nikolce Murgovski – Chalmers University of Technology
tel: +46 72 170 51 72
mail: nikolce.murgovski@chalmers.se

Erik Börve – Volvo GTT / Chalmers University of Technology
Mail: erik.borve@volvo.com

 

...

Job Category:  Technology Engineering

Organization:  Group Trucks Technology

Travel Required:  No Travel Required

Requisition ID:  14846