CS MSc Thesis Presentation 15 September 2026

Tuesday, 15 September there will be a master thesis presentation in Computer Science at Lund University, Faculty of Engineering.

The presentation will take place in E:4130 (Lucas).

Note to potential opponents: Register as an opponent to the presentation of your choice by sending an email to the examiner for that presentation (firstname [dot] lastname [at] cs [dot] lth [dot] se). Do not forget to specify the presentation you register for! Note that the number of opponents may be limited (often to two), so you might be forced to choose another presentation if you register too late. Registrations are individual, just as the oppositions are! More instructions for opponents are found here on the LTH thesis project page.

13:00-14:00 in E:4130 (Lucas)

• Presenter: Zhehao Chen
• Title: Transformer-based Normalizing Flow for Robot Arm Control---RoboTarFlow
• Examiner: Davide Tateo
• Supervisors: Volker Krueger (LTH)

This thesis investigates whether a transformer-based autoregressive normalizing flow can serve as a visuomotor policy for bimanual robot manipulation. We adapt the TarFlow image-generation architecture so that each token represents a timestep of a joint command rather than an image patch, and condition the flow on camera observations via cross-attention. The resulting policy, RoboTarFlow, is evaluated on all 50 RoboTwin 2.0 Easy tasks against two baselines: NF-P and Diffusion Policy. Ablation studies isolate the design choices that matter: the prediction horizon and frame stride dominate, followed by the vision encoder, where a self-supervised transformer backbone DINOv2 clearly outperforms ResNet18, and by the training action noise. The results indicate that autoregressive normalizing flows are a viable alternative to diffusion-based policies, reaching competitive performance while retaining exact likelihood computation.