30 Credit - Self-Supervised Learning using Foundation Models

Introduction 
Thesis work is an excellent way to get closer to Scania and build relationships for the future. Many of today's employees began their Scania career with their degree project. 

Background  
Scania is now undergoing a transformation from being a supplier of trucks, buses and engines to a supplier of complete and sustainable transport solutions. 

This thesis work will lie under the supervision of the research group EEARP, which develops the algorithms that are used in the scene perception for autonomous driving. 

Objective  
In recent years, advancements in 3D object detection have become pivotal in the development of autonomous systems. A major challenge in this domain is the reliance on large, manually labelled datasets. This manual annotation process is both time-consuming and expensive, requiring human expertise to create precise 3D bounding boxes for various object categories. As new sensor technologies and datasets evolve, the need for constantly updating these labels compounds the challenge. To address this, autolabeling techniques have gained significant attention. These methods aim to minimize human intervention while maintaining or even improving the accuracy and efficiency of object detection systems. 

Among existing methods, some leverage purely 3D sensor data, such as LiDAR, to generate object proposals by analysing motion patterns within point cloud sequences. By focusing on data points that are mobile, the method can accurately identify and label both static and dynamic objects over time, refining the detection through continuous self-training based on object trajectories. Another class of methods use the visual appearance of objects in the scene to cluster similar objects together, improving the precision of labelling even for static instances, thereby avoiding multiple rounds of training. Foundation models have the potential to improve the speed and accuracy of these processes further by avoiding expensive round of trainings and providing feature rich prior for processing. At the same time, traditional optimization based approaches could enable robustness to occlusions or sensor sparsity. 

Job description  
We are seeking a talented student to explore development of robust autolabeling pipeline on multimodal data, leveraging foundation models as well as traditional optimization approaches. Your tasks will include (but not be limited to): 

• Conducting literature survey and on methods for autolabeling, 3D object discovery and self-supervised learning for autonomous driving scenarios. 

• Benchmarking and training existing methods on available open datasets. 

• Finding and addressing gaps in existing methodologies. 

• Document findings in form of a report and presentations. 

The successful applicant will have the opportunity to apply state-of-the-art methods to real world scenarios and gain hands-on experience on our in-house rich datasets, the latest sensors, computing platforms, and Scania’s concept autonomous vehicles. The applicant will also collaborate with dynamic and experienced researchers and developers working at Scania’s Autonomous Transport Solutions Pre-Development & Research department. 

Requirements:  

• Studying M.Sc. in Machine Learning, Systems Controls & Robotics, Computer Science, Engineering Physics or similar programs. 

• Experience with Python and version control (eg. Git). 

• Experience with deep learning frameworks eg. PyTorch. 

• Able to work in a diverse environment and communicate effectively in English 

• Excellent problem-solving skills and the ability to work independently 

Preferred: 

• Interest in 3D computer vision, deep learning and foundation model research. 

• Interest in publishing or writing a research paper for computer vision conferences. 

• Experience handling large outdoor datasets, eg. NuScenes, Waymo, Argoverse2 etc. 

• Experience with frameworks eg. mmdetection3d, openPCDet. 

Number of students: 1 

Time plan  
The project is planned for 20 weeks and can be started any time in early Spring 2025. Applicants will be assessed on a continuous basis until the position is filled. 

Contact persons and supervisors 
Ajinkya Khoche, PhD Candidate, Scania Autonomous Transport Solutions, ajinkya.khoche@scania.com  

Application: 
Your application must include a CV, personal letter and transcript of grades  

A background check might be conducted for this position. We are conducting interviews continuously and may close the recruitment earlier than the date specified.