Master thesis - Robometer for Autonomous Driving Data Selection

30 credits - Robometer for Autonomous Driving Data Selection 

Background 

The development of models for autonomous driving rely on large-scale driving datasets. While the availability of data has grown substantially, not all recorded driving data contributes equally to model performance. Large portions of driving logs consist of repetitive or uninformative scenarios, whereas rare, safety-critical, or highly interactive situations often provide significantly greater learning value. Consequently, identifying which data should be prioritized during training has become an important research problem in autonomous driving. Current data selection strategies typically rely on heuristic measures such as scenario rarity, uncertainty estimates, diversity metrics, or manually designed filters. More recently, learned data valuation methods have been proposed to estimate the usefulness of individual training samples based on their contribution to downstream model performance. 

Robometer [1] is a recently proposed foundation reward model for robotics that learns to assess the quality and progress of robot trajectories from large-scale offline data. Rather than evaluating individual observations, Robometer assigns meaningful scores to trajectories based on learned notions of task completion and execution quality. The model has demonstrated strong generalization across robotic manipulation tasks and has been shown to outperform heuristic reward functions in applications such as reinforcement learning, imitation learning, and data retrieval. 

Although Robometer has been developed for robotic manipulation, its underlying principle of learning trajectory quality is highly relevant for autonomous driving. Driving datasets naturally consist of trajectories with varying levels of complexity, interaction, and execution quality. A learned trajectory scoring model could therefore provide an informative signal for selecting the most valuable training examples, filtering redundant data, or constructing curricula for training driving foundation models. 

Zero-shot scoring of an autonomous driving scenario with RoboMeter for the task prompt: Safely traverse the intersection when the traffic light turns green. 

Objective 

The goal of this thesis is to investigate whether Robometer can be adapted and deployed as a trajectory quality metric for data selection in autonomous driving. 

This includes: 

• Investigating how Robometer can be applied to autonomous driving trajectories and fine- 
  tuning of the model. 

• Developing a data-selection pipeline based on Robometer trajectory scores. 

• Comparing Robometer with existing data-selection strategies. 

• Evaluate whether Robometer-selected datasets improve downstream autonomous driving mod- 
  els in terms of performance, robustness, and data efficiency. 

The Project Offers
 

The student will work on a research problem combining machine learning, autonomous driving and multimodal models. The student will receive supervision from TRATON and have opportunities to interact with researchers and engineers working on machine learning and autonomous driving. 

Who are we looking for? 

• We are looking for a Master’s student in computer science, machine learning, robotics, engineering physics, electrical engineering, or a related field. 

• A suitable candidate should have: 

• strong programming skills, preferably in Python and PyTorch; 

• knowledge of machine learning and deep learning; 

• an interest in autonomous driving, computer vision, transformers, multimodal models, or explainable AI; 

• experience with vision-language models, natural-language processing, model interpretability, or autonomous-driving datasets is beneficial but not required; and 

• motivation to combine scientific investigation with practical implementation. 

The planned thesis start is January 2027. 

Number of students: 1 

Start date for the thesis work: [To be agreed] 

Estimated time required: 20 weeks, full time (30 credits) 

Contact persons and supervisors 

Caroline Skoglund, Industrial PhD Student (KTH Robotics, Perception and Learning) caroline.skoglund@scania.com 

Cristina Cipriani, Research & Development Engineer in Autonomous Motion, cristina.cipriani@scania.com 

Hiring Manager: Magnus Granström, magnus.granstrom@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. 

References 

1. Robometer: Scaling General-Purpose Robotic Reward Models via Trajectory Comparisons