About

I am a PhD student at the Institute for Automotive Engineering (RWTH Aachen University), advised by Prof. Lutz Eckstein. In my PhD thesis, I was working on Trustworthy AI for perception models, i.e. training explainable, robust, and probabilistic models. Currently, I am focussing on foundation models / VLAs for robotics and automated driving.

My background lies in Computational Engineering Science and I graduated with distinction from RWTH Aachen University with a study abroad semester in the US.

Outside of research, I enjoy traveling to new places, photography, building software tools and playing around with off-grid communication.

GitHub Google Scholar LinkedIn Hugging Face
Portrait of Till Beemelmanns
Till Beemelmanns, PhD Researcher

News

May 2026"Towards Trustworthy and Explainable AI for Perception Models: From Concept to Prototype Vehicle Deployment" accepted at ITSC 2026
Feb 2026"Query2Uncertainty" accepted at CVPR 2026
May 2025"OCCUQ" accepted at ICRA 2025
Jan 2024Won the GME Award for MOOC "Automated and Connected Driving Challenges"
Jan 2024"MultiCorrupt" accepted at IV 2024
Nov 2023"Explainable Multi-Camera 3D Object Detection with Transformer-Based Saliency Maps" accepted at ML4AD Workshop at NeurIPS 2023
Nov 2021"3D Point Cloud Compression with Recurrent Neural Network and Image Compression Methods" accepted at IEEE IV 2022

Publications

Query2Uncertainty

Query2Uncertainty: Robust Uncertainty Quantification and Calibration for 3D Object Detection under Distribution Shift

Till Beemelmanns, Alexey Nekrasov, Stefan Vilceanu, Jonas Steinhaus, Timo Woopen, Bastian Leibe, Lutz Eckstein

IEEE/CVF CVPR — 2026

arXiv PDF Project BibTeX
Towards Trustworthy and Explainable AI for Perception Models

Towards Trustworthy and Explainable AI for Perception Models: From Concept to Prototype Vehicle Deployment

Till Beemelmanns, Shayan Sharifi, Manas Mehrotra, Ayushman Choudhuri, Lutz Eckstein

IEEE ITSC — 2026

arXiv PDF Project BibTeX
karl. — A Research Vehicle for Automated and Connected Driving

karl. — A Research Vehicle for Automated and Connected Driving

Jean-Pierre Busch, Lukas Ostendorf, Guido Linden, Lennart Reiher, Till Beemelmanns, Bastian Lampe, Timo Woopen, Lutz Eckstein

IEEE IV — 2026

arXiv BibTeX
OCCUQ

OCCUQ: Exploring Efficient Uncertainty Quantification for 3D Occupancy Prediction

Severin Heidrich*, Till Beemelmanns*, Alexey Nekrasov*, Bastian Leibe, Lutz Eckstein *equal contribution

IEEE ICRA — 2025

arXiv Code BibTeX
MultiCorrupt

MultiCorrupt: A Multi-Modal Robustness Dataset and Benchmark of LiDAR-Camera Fusion for 3D Object Detection

Till Beemelmanns, Quan Zhang, Christian Geller, Lutz Eckstein

IEEE IV — 2024

arXiv Code Dataset BibTeX
Explainable Multi-Camera 3D Object Detection

Explainable Multi-Camera 3D Object Detection with Transformer-Based Saliency Maps

Till Beemelmanns, Wassim Zahr, Lutz Eckstein

ML4AD Workshop, NeurIPS — 2023

arXiv PDF BibTeX
3D Point Cloud Compression

3D Point Cloud Compression with Recurrent Neural Network and Image Compression Methods

Till Beemelmanns, Yuchen Tao, Bastian Lampe, Lennart Reiher, Raphael van Kempen, Timo Woopen, Lutz Eckstein

IEEE IV — 2022

arXiv Code BibTeX
Self-learning Trajectory Prediction with Recurrent Neural Networks

Self-learning Trajectory Prediction with Recurrent Neural Networks at Intelligent Intersections

Julian Bock, Till Beemelmanns, Markus Klösges, Jens Kotte

VEHITS — 2017

PDF BibTeX

Other Publications

Carlos: An Open, Modular, and Scalable Simulation Framework for the Development and Testing of Software for C-ITS

Christian Geller, Benedikt Haas, Amarin Kloeker, Jona Hermens, Bastian Lampe, Till Beemelmanns, Lutz Eckstein

IEEE IV — 2024 arXiv

Enabling Connectivity for Automated Mobility: A Novel MQTT-based Interface Evaluated in a 5G Case Study on Edge-Cloud LiDAR Object Detection

Lennart Reiher, Bastian Lampe, Timo Woopen, Raphael Van Kempen, Till Beemelmanns, Lutz Eckstein

IEEE ICECCME — 2022 arXiv

Data-driven Occupancy Grid Mapping using Synthetic and Real-World Data

Raphael van Kempen, Bastian Lampe, Lennart Reiher, Timo Woopen, Till Beemelmanns, Lutz Eckstein

IEEE ICECCME — 2022 arXiv

Automation of the UNICARagil Vehicles

Michael Buchholz, Fabian Gies, Andreas Danzer, Matti Henning, Charlotte Hermann, Manuel Herzog, Markus Horn, Markus Schön, Nils Rexin, Klaus Dietmayer, Carlos Fernandez, Johannes Janosovits, Danial Kamran, Christian Kinzig, Martin Lauer, Eduardo Molinos, Christoph Stiller, Lingguang Wang, Stefan Ackermann, Tobias Homolla, Hermann Winner, Grischa Gottschalg, Stefan Leinen, Matthias Becker, Johannes Feiler, Simon Hoffmann, Frank Diermeyer, Bastian Lampe, Till Beemelmanns, Raphael van Kempen, Timo Woopen, Lutz Eckstein, Nicolai Voget, Dieter Moormann, Inga Jatzkowski, Torben Stolte, Markus Maurer, Jürgen Graf

29th Aachen Colloquium Sustainable Mobility — 2020 PDF

Theses & Projects

Project: Website Fingerprinting and Traffic Labeling with Deep Neural Networks

Till Beemelmanns, Adam Weckle

Michigan State University, 825 Computer Security — November 2018 PDF

Project: Convolutional Neural Network and Recurrent Neural Network for Earthquake Detection and Localization

Till Beemelmanns

Michigan State University, 801 Computational Modeling — December 2018 PDF

Seminar Thesis: Surrounding Object Trajectory Prediction with Recurrent Neural Networks

Till Beemelmanns

RWTH Aachen University — December 2017 PDF

Bachelor Thesis: Continuously Learning Prediction of Pedestrian Movements at Intersections with Recurrent Neural Networks

Till Beemelmanns

RWTH Aachen University — March 2017 PDF

Projects

A ROS 2 node for Depth Anything V3 depth estimation using TensorRT for real-time inference. This node subscribes to camera image and camera info topics and publishes directly both, a metric depth image and PointCloud2 point cloud.

ros2-depth-anything-v3-trt

ROS2 C++ ONNX TensorRT
ros2_calib is a multi-sensor calibration tool for ROS 2 that provides intuitive graphical interfaces for performing precise extrinsic calibration between different sensor types. The tool supports both LiDAR-to-Camera and LiDAR-to-LiDAR calibration workflows. Built with PySide6, it operates on recorded rosbag data without requiring a live ROS 2 environment. It supports reading /tf_static transforms from rosbags and allows users to quickly calibrate and export the resulting transformation directly into URDF format.

ros2_calib

Python
ros2_tui is a terminal user interface for ROS 2 that shows live topic details, message frequency, and delay statistics, with an echo pane for inspecting message contents.

ros2_tui

Rust ROS2