Master Thesis - In-Vehicle VLA Deployment for L4 Autonomous Driving
Göteborg, SE, 417 15 Göteborg, SE, 405 08
Transport is at the core of modern society. Imagine using your expertise to shape sustainable transport and infrastructure solutions for the future. If you seek to make a difference on a global scale, working with next-gen technologies and the sharpest collaborative teams, then we could be a perfect match.
What you will do
Master Thesis proposal: In-Vehicle End-to-End Learning with Vision-Language-Action Models for L4 Autonomous Driving: Architecture Design and Deployment
Autonomous driving seeks to enable vehicles to perceive and understand their surroundings and make driving decisions without human intervention. Recent end-to-end approaches increasingly replace hand-engineered pipelines with learned policies that directly map sensor inputs to actions, with emerging VLA-based systems such as NVIDIAs Alpamayo exploring multimodal reasoning and direct trajectory generation for autonomous driving. While impressive performance has been demonstrated, many questions remain concerning real-time performance and safety redundancy.
This thesis focuses on bringing a VLA driving model into a real vehicle. Using any of the open Alpamayo models on a 128 GB unified-memory (UMA) computer as the compute core. You will propose a proof-of-concept in-vehicle system architecture for the Volvo FH truck of the Volvo–Chalmers ReVeRe research-vehicle lab, benchmarked against NVIDIA's production DRIVE AGX Thor reference.
Because the UMA desktop (e.g. DGX Spark) and DRIVE AGX Thor share the same Blackwell + CUDA/TensorRT architecture, a policy prepared on the bench maps onto the in-vehicle unit. The architecture must handle real truck sensing and the wider articulated swept path and larger negotiation margins that distinguish a combination from a car [1], always within a supervised safety envelope (safety driver plus an independent redundant layer). The target is a supervised highway-pilot demonstration, including the exit-ramp lane-change case on the Volvo FH at AstaZero in May–June 2027.
The main objectives of this thesis are:
- PoC system architecture: Propose an in-vehicle architecture that runs a VLA model on a 128 GB UMA computer in the ReVeRe Volvo FH, mapped against the DRIVE AGX Thor reference (sensing, compute, DriveOS/TensorRT, safety envelope).
- Bench & hardware-in-the-loop: Bring up the chosen Alpamayo model on the UMA computer and validate real-time latency and update-rate budgets against Thor targets.
- Vehicle integration & demonstration: Integrate on the Volvo FH (ReVeRe) and prepare a supervised highway-pilot demonstration at AstaZero (May–June 2027), including the multiple-lane-change-to-exit scenario.
Who are you?
This project combines embedded/real-time systems, model optimization and vehicle integration (C++/Python, TensorRT). The work will be carried out at Volvo Group, Göteborg, within the Volvo–Chalmers ReVeRe collaboration, and is recommended for two (2) students with a strong background in embedded/real-time systems and Python/C++.
Thesis team: this is one of three linked master theses – (1) using the car-trained models as-is by prompting, (2) closed-loop reinforcement-learning post-training, and (3) in-vehicle deployment – forming a single thesis team that meets regularly and shares simulation infrastructure and results.
Contact persons
Erik Börve – Volvo TTI, email: erik.borve@volvo.com
Markus G. – Volvo TTI (co-supervisor)
Bibliography
[1] P. Nilsson, "Traffic Situation Management for Driving Automation of Articulated Heavy Road Transports – From driver behaviour towards highway autopilot," PhD thesis, Chalmers University of Technology, Göteborg, 2017.
[2] NVIDIA DRIVE AGX / DRIVE AGX Thor (in-vehicle compute + DriveOS): https://developer.nvidia.com/drive/agx ; https://developer.nvidia.com/drive/ecosystem-thor
[3] NVIDIA Alpamayo & AlpaSim: https://www.nvidia.com/en-us/solutions/autonomous-vehicles/alpamayo/ ; https://github.com/NVlabs/alpasim
[4] NVIDIA DGX Spark (128 GB unified-memory desktop): https://www.nvidia.com/en-us/products/workstations/dgx-spark/
Last application date: November 1st, 2026.
Location: Gothenburg, Sweden.
Time schedule: Jan 2027 - Jun 2027.
We value your data privacy and therefore do not accept applications via mail.
Who we are and what we believe in
We are committed to shaping the future landscape of efficient, safe, and sustainable transport solutions. Fulfilling our mission creates countless career opportunities for talents across the group’s leading brands and entities.
Applying to this job offers you the opportunity to join Volvo Group. Every day, you will be working with some of the sharpest and most creative brains in our field to be able to leave our society in better shape for the next generation. We are passionate about what we do, and we thrive on teamwork. We are almost 100,000 people united around the world by a culture of care, inclusiveness, and empowerment.
Group Trucks Technology are seeking talents to help design sustainable transportation solutions for the future. As part of our team, you’ll help us by engineering exciting next-gen technologies and contribute to projects that determine new, sustainable solutions. Bring your love of developing systems, working collaboratively, and your advanced skills to a place where you can make an impact. Join our design shift that leaves society in good shape for the next generation.