Holger Caesar
Holger Caesar is an Assistant Professor at TU Delft who builds the datasets and methods behind autonomous-vehicle perception. After studies in Germany and Switzerland, he earned a computer vision PhD at the University of Edinburgh under Vittorio Ferrari, with a 2018 thesis that includes the COCO-Stuff dataset. He coauthored PointPillars in 2019 and developed the nuScenes dataset and the nuPlan planning benchmark at the autonomous-vehicle company Motional, where he was a Principal Research Scientist. In April 2022 he joined the Intelligent Vehicles group at TU Delft, and he became an ELLIS Scholar in 2024. Since March 2026, he is Co-Director of the ELLIS Delft Unit.
Timeline
-
By 2018
A computer vision PhD
He studies at KIT Karlsruhe, EPF Lausanne and ETH Zurich, then earns a PhD in computer vision at the University of Edinburgh under Vittorio Ferrari. His 2018 thesis includes COCO-Stuff, which adds pixel-wise annotations for 91 stuff classes to all 164K images of the COCO 2017 dataset.
-
2019
PointPillars for lidar detection
He coauthors PointPillars with Alex H. Lang, Sourabh Vora and others, an encoder that organizes lidar point clouds in vertical columns for object detection. The method runs at 62 Hz, and the paper is accepted at CVPR 2019.
-
2019–2020
The nuScenes dataset
He is first author of nuScenes, the first dataset to carry the full autonomous vehicle sensor suite of 6 cameras, 5 radars and 1 lidar. The paper reaches arXiv in March 2019 and CVPR 2020, and Motional publishes it in May 2020. He develops and leads the dataset during his time at Motional, formerly nuTonomy.
-
2021
The nuPlan benchmark
In June 2021, he is first author of the nuPlan paper, which proposes the first closed-loop ML-based planning benchmark for autonomous driving. It offers 1500 hours of human driving data from Boston, Pittsburgh, Las Vegas and Singapore.
-
April 2022
Assistant Professor at TU Delft
In April 2022, he starts as Assistant Professor in the Intelligent Vehicles group at TU Delft, after serving as Principal Research Scientist at Motional, where he started 3 teams with 20+ members for data annotation, autolabeling and data mining. His research covers perception and prediction for autonomous vehicles, with a focus on scalable learning and annotation.
-
February 2024
Perception with almost no labels
In an ELLIS Delft talk in February 2024, he describes nuScenes and PointPillars as the first generation of perception systems and nuPlan as the second. He envisions a third generation that needs little to no human labeling, using active learning, object discovery, self-supervised learning and foundation models.
-
October 2024
An ELLIS Scholar
He becomes an ELLIS Scholar in October 2024 and joins the board of ELLIS Delft. TU Delft names him among four new ELLIS Scholars, with Wendelin Böhmer, Emir Demirović and Xucong Zhang.
-
Since March 2026
Co-director of ELLIS Delft
In March 2026, he becomes Co-Director of the ELLIS Delft Unit, which lists him and Frans A. Oliehoek as Unit Directors. He is an Assistant Professor in the Intelligent Vehicles group at TU Delft.
Career
- Since 2024European Laboratory for Learning and Intelligent SystemsBoard Member, Co-Director
- Since 2022Delft University of TechnologyAssistant Professor
- Until 2022MotionalPrincipal Research Scientist
Education
Sources and credits
Photo: We have no published photo of Holger Caesar yet. Send us a free-licensed photo.
- Dr. H. (Holger) Caesar
- Restoring the balance between stuff and things in scene understanding
- COCO-Stuff: Thing and Stuff Classes in Context
- PointPillars: Fast Encoders for Object Detection from Point Clouds
- nuScenes: A multimodal dataset for autonomous driving
- nuScenes: A multimodal dataset for autonomous driving - Motional
- Holger Caesar's Homepage - Media
- NuPlan: A closed-loop ML-based planning benchmark for autonomous vehicles
- Holger Caesar's Homepage
- ELLIS Delft Talk by Holger Ceasar: Building machine perception systems with (almost) no labels
- 9 researchers from ELLIS Unit Delft honoured with Fellows or Scholars title
- ELLIS Unit Delft

