Speakers
Prof. Dr. Christoph Stiller
The Institute of Measurement and Control Systems at the Karlsruhe Institute of Technology (KIT), lead by Prof. Dr.-Ing. Christoph Stiller, and the Research Center for Information Technology (FZI) jointly research various topics along the full stack of autonomous driving. With their autonomously driving research vehicles, they successfully participated in various challenges, such as the DARPA Urban Challenge or the Grand Cooperative Driving Challenges. Moreover, they are known for widely used research datasets, such as the KITTI Vision Benchmark Suite and the INTERACTION Dataset. Finally, with Lanelet2, they developed one of the leading open-source HD map frameworks.
Localization and Map Integrity for Map Learning
Frank Bieder, Haohao Hu, Isabell Hofsetter, and Jan-Hendrik Pauls
For safe and comfortable autonomous driving, maps are crucial as they can compensate sensor and processing weaknesses. To safely use map information, two requirements need to be met: Firstly, the autonomous vehicle needs to know where it is and how certain this estimate is (localization integrity). Secondly, it needs to verify which parts of the map are still up-to-date (map integrity).
With accurate 6D localization, maps can also act as ground truth that scales up extremely well (map learning). Hence, in the age of machine learning, localization and map integrity become even more significant as they assess the quality of this very ground truth.
Dr. Mathieu Joerger obtained a Master in Mechatronics from INSA Strasbourg, France, in 2002, and a M.S. and Ph.D. in Aerospace Engineering from the Illinois Institute of Technology, Chicago, in 2002, and 2009, respectively. He is the 2009 recipient of the Institute of Navigation (ION) Bradford Parkinson award, and the 2014 recipient of the ION Early Achievement Award. He is Technical Editor of Navigation for the Institute of Electrical and Electronics Engineers (IEEE) Transactions on Aerospace and Electronic Systems. Dr. Joerger is currently assistant professor at Virginia Tech, working on multi-sensor integration for ground vehicle navigation, and on fault detection and exclusion for civilian aviation applications.
Dr. Bisheng Yang is the professor in Geomatics, the State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing (LIESMARS), Wuhan University. His research interests are Laser scanning and UAV photogrammetry; Semantic parsing and modeling of point clouds; Three-dimensional geographic information acquisition and analysis; Intelligent location based services.
Real time semantic understanding in dynamic environments with 3D Lidar SLAM
With the development of UAV and robots, light-weighted, low-cost, and flexible 3D Lidar SLAM based systems are attracting intensive attention both from academic and industrial communities. High flexibility and low-cost wearable 3D Lidar SLAM systems are urgently required for wide purposes, e.g., difficult scenarios (e.g., GNSS-denied areas), autonomous driving. This talk will give more details about the performance evaluation of a 3D Lidar SLAM based system that is designed and implemented by the research group, including the system calibration, fusion of Lidar/IMU for accurate attitude estimation, real time point clouds semantic understanding, and several case studies.
Dr. Naoki Akai has been worked on the autonomous navigation field for robots and cars. His current research objective is to establish a method to guarantee safety of autonomous navigation. Particularly, I'm focusing on the ego-vehicle localization. To this end, he is mainly studying probabilistic modeling, deep learning, and topological data analysis.
Probabilistic Approach to Reliable Localization
To realize reliable localization, (1) robust localization in dynamic environments, (2) precis failure detection, and (3) immediate failure recovery must be achieved. In this talk, I present a probabilistic approach to reliable localization that simultaneously achieves above three functions. The approach is formulated as a joint posterior estimation problem over a robot pose, sensor measurement class, and localization state. In addition, the approach includes effective fusion of pose tracking with global localization and it enables to achieve immediate and stable recovery from localization failure. I focus on 2D-LiDAR-based localization and show that the presented approach contributes to realize reliable localization.
Dr. Chen Zhu is a senior research fellow and the head of Visual and Terrestrial Augmentation group at the Institute of Communications and Navigation, German Aerospace Center (DLR). He received his Ph.D. degree (Dr.-Ing.) and Master’s degree (M.Sc.) from Technical University of Munich, Germany and his Bachelor’s degree from Tsinghua University, Beijing, China. He is interested in the research fields of visual navigation, multi-sensor fusion, and robotic swarm navigation, currently focusing on the system integrity.
Integrity of visual navigation: an important aspect for safety-critical applications
Dr. Boubeker Belabbas
Dr. Boubeker Belabbas is a senior navigation expert at BOSCH Group working in the field of autonomous driving. He has more than 20 years experience in GNSS and its augmentations (Satellite and Ground Based Augmentation Systems SBAS and GBAS, multi-sensor fusion) and was leading the Navigation Integrity Group at the German Aerospace Center (DLR) since 2011 and was project manager of the first worldwide test demonstrating GBAS automatic landing of aircraft using single frequency GPS i.e. GBAS Approach Service Type D.
A modular approach of performance based localization for autonomous driving
Dr. Philippe Bonnifait is a professor at the Université de Technologie de Compiègne, Alliance Sorbonne Université Research lab: Heudiasyc UMR CNRS 7253
Localization Integrity for Intelligent Vehicles: How and for what?
In theory, a vehicle could navigate with only a perception system that would understand everything about its surroundings. Moreover, we often navigate more efficiently on roads we have already traveled thanks to the geospatial knowledge of the environment.
Localization and maps are a way to simplify navigation tasks, for autonomous vehicles or advanced driving assistance. Indeed, localization associated with a High Definition (HD) map facilitates the understanding of the driving situation, allowing to focus attention on areas of interest, to identify interacting road users and to anticipate the situation based on prediction mechanisms. Global localization is also a way to increase perception beyond the sensors' field of view, in hidden areas or in degraded conditions thanks to the real-time exchange of data between road users in a cooperative system.
Localization being used in many autonomous navigation tasks, it has to be non-misleading. Integrity monitoring in real-time becomes a key issue as well as defining its requirements which are task-dependent (and very diverse). This is a challenge that the scientific community faces when it comes to defining the different levels of requirements.
In this talk, I will present a multi-sensor localization system studied in the lab using GNSS, odometry and exteroceptive sensors (Lidar and camera) able of detecting and associating geo-referenced features in HD maps such as road markings or traffic signs.
I will also present results we have recently obtained in the lab regarding the management of events that can lead to integrity losses and misleading localization information for autonomous vehicles. A particular focus will be on the management of erroneous sensor data, map errors, and association problems.
Finally, the question of the choice of a TIR (Target Integrity Risk) for the calculation of a Protection Level (PL) will be addressed in order to initiate discussions with the workshop participants.