Nils Blank

I am a PhD student in the Intuitive Robots Lab (IRL) at the Karlsruhe Institute of Technology (KIT), Germany. My research focuses on Imitation Learning and Foundation Models for Human-Robot-Interaction. I am supervised by Rudolf Lioutikov. I obtained my Master's Degree in Computer Science at the KIT. During my studies I interned at SAP SE and IONOS.

Research

My research focuses on Foundation Models and their applications in Robotics. In particular, I explore how we can employ Foundation models robustly and reliably in challenging robotic scenarios. Furthermore, my research focuses on goal driven explainability and how we can leverage foundation models for improved human-robot interaction.

Publications

Google Scholar

Overview of SPARC's spatial annotation pipeline

CoRL 2026

SPARC: Reliable Spatial Annotations from Robot Demonstrations at Scale

Nils Blank, Paul Mattes, Maximilian Xiling Li, Jakub Suliga, Thomas Roth, Moritz Reuss, Pankhuri Vanjani, Rudolf Lioutikov

Best Paper Award · Lab2Production Workshop at RSS 2026

SPARC turns robot demonstrations into spatial annotations with reliability scores, enabling control over label quality without human review. It retains three times more samples at high precision on IA-Bench and improves spatial grounding and robot manipulation.

Structured scene graph representation used by SIR

CVPR 2026

SIR: Structured Image Representations for Explainable Robot Learning

Paul Mattes, Jan Schwab, Jens Bosch, Maximilian Xiling Li, Nils Blank, Minh-Trung Tang, Moritz Haberland, Rudolf Lioutikov

SIR learns sparse, task-relevant scene graphs that make robot policies interpretable. On RoboCasa, it improves success from 14.81% to 19.5% while making policies more robust to distractors and exposing dataset biases.

BEAST tokenizer for B-spline action sequences

NeurIPS 2025

BEAST: Efficient Tokenization of B-Splines Encoded Action Sequences for Imitation Learning

Hongyi Zhou, Weiran Liao, Xi Huang, Yucheng Tang, Fabian Otto, Xiaogang Jia, Xinkai Jiang, Simon Hilber, Ge Li, Qian Wang, Ömer Erdinç Yağmurlu, Nils Blank, Moritz Reuss, Rudolf Lioutikov

BEAST encodes robot action sequences into compact B-spline tokens without separate tokenizer training. Uniform-length tokens support fast parallel decoding while producing smooth, continuous trajectories.

Foundation-model pipeline for labeling robot demonstrations

CoRL 2024

Scaling Robot Policy Learning via Zero-Shot Labeling with Foundation Models

Nils Blank, Moritz Reuss, Marcel Rühle, Ömer Erdinç Yağmurlu, Fabian Wenzel, Oier Mees, Rudolf Lioutikov

NILS combines pretrained foundation models to segment and label unstructured robot demonstrations with natural language, without human intervention. We annotate over 115,000 trajectories from more than 430 hours of robot data.