Jürgen Sturm is a
Germansoftware engineer, entrepreneur and academic. He is a Senior Staff Software Engineering Manager at Intrinsic, where he works on developing a robot
SDK aimed at facilitating and reducing the cost of integrating AI-/ML-powered robots into industrial manufacturing processes.[1]
Sturm earned his bachelor's and master's degrees in Artificial Intelligence from the
University of Amsterdam in 2006, followed by a PhD in
Robotics from the
University of Freiburg, with his later thesis published as a book in 2013.[7] From 2011 to 2014, he served as a Postdoctoral Researcher in the Computer Vision group at the Technical University of Munich (TUM), where he worked on real-time camera tracking and
3D person scanning methods. Concurrently, he began his academic career, delivering lectures at TUM and teaching an online course at
EdX in 2012 and 2013.[8]
Career
At TUM, Sturm developed a
3D reconstruction algorithm enabling
3D scanning of a person for printing as a small figure,[9] leading to him co-founding the 3D scanning startup FabliTec in 2013, where he served as CEO until 2015.[10] In 2014, he joined
Metaio as a Senior Software Developer and Team Lead.[11] Subsequently, he was appointed Senior Software Engineer and Tech Lead Manager at
Google.[12] leading to multiple patents.[13][14] He assumed the position of an Engineering Manager at Intrinsic in 2019.[1]
Research
Sturm has contributed to the field of engineering by studying robotics, machine intelligence and
machine perception, holding several patents for his developments in RGB-D cameras and 3D mapping techniques.[2]
RGB-D SLAM
Sturm has researched and worked on RGB-D cameras throughout his career. In a collaborative effort, he presented a benchmark for RGB-D
SLAM systems, offering high-quality image sequences with accurate ground truth camera poses, diverse scenes, and automatic evaluation tools accessible through a dedicated website.[15] He also proposed a dense visual SLAM method for RGB-D cameras, alongside
Daniel Cremers and
Wolfram Burgard, improving pose accuracy by minimizing errors.[16] Additionally, he showcased an RGB-D camera SLAM system for the
Microsoft Kinect, assessing its accuracy, robustness, and speed across different indoor scenarios and offering it as open-source software.[17]
3D mapping
Sturm's work on 3D mapping focused on reconstruction and improving techniques for precision. Alongside colleagues, he demonstrated a mapping system using RGB-D cameras for accurate 3-D mapping.[18] He also introduced a real-time mapping system for RGB-D images using an octree structure to update a textured triangle mesh, enabling efficient memory usage for mobile or flying robots,[19] as well as a new real-time visual odometry method for monocular cameras, achieving superior accuracy and speed by continuously estimating a semi-dense inverse depth map.[20] Furthermore, he presented a 3D reconstruction algorithm based on Truncated Signed Distance Functions (TSDF), addressing the challenge of representing dynamic environments for robots, with a focus on continuous refinement of static maps and robust scene differencing.[21]
In a joint research effort, Sturm proposed a graph-based method to calibrate sensor suites for accurate direct georeferencing of images from small unmanned aerial systems, addressing static offsets and in-flight calibration of intrinsic camera parameters.[22]
3D perception
Sturm has been involved in the development of models for 3D perception and scanning as well. He presented ScanComplete, a data-driven method using a generative 3D CNN model to predict complete 3D models with semantic labels from incomplete scans.[23] In addition, he revealed a real-time RGB-D scene understanding method for mobile devices, combining incremental reconstruction, geometric segmentation, and semantic labeling.[24]
Awards and honors
2011 – Best Dissertation Award, European Coordinating Committee of Artificial Intelligence (ECCAI)[3]
2011 – Wolfgang-Gentner-Award for an Outstanding PhD Thesis, University of Freiburg[4]
2012 – Best Research Paper Award, Unmanned Aerial Vehicle in Geomatics
2012, 2013 – TeachInf Best Lecture Award, Technical University of Munich[5]
Approaches to Probabilistic Model Learning for Mobile Manipulation Robots (2013) ISBN 978-3642371592
Selected articles
Endres, F., Hess, J., Engelhard, N., Sturm, J., Cremers, D., & Burgard, W. (2012, May). An evaluation of the RGB-D SLAM system. In 2012 IEEE international conference on robotics and automation (pp. 1691-1696). IEEE.
Sturm, J., Engelhard, N., Endres, F., Burgard, W., & Cremers, D. (2012, October). A benchmark for the evaluation of RGB-D SLAM systems. In 2012 IEEE/RSJ international conference on intelligent robots and systems (pp. 573-580). IEEE.
Kerl, C., Sturm, J., & Cremers, D. (2013, November). Dense visual SLAM for RGB-D cameras. In 2013 IEEE/RSJ International Conference on Intelligent Robots and Systems (pp. 2100-2106). IEEE.
Rethage, D., Wald, J., Sturm, J., Navab, N., & Tombari, F. (2018). Fully-convolutional point networks for large-scale point clouds. In Proceedings of the European Conference on Computer Vision (ECCV) (pp. 596-611).
Dai, A., Ritchie, D., Bokeloh, M., Reed, S., Sturm, J., & Nießner, M. (2018). Scancomplete: Large-scale scene completion and semantic segmentation for 3d scans. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 4578-4587).