Workshop on Traffic Infrastructure Mapping and Automated Damage Assessment Systems
In recent years, the automated digital capture of traffic infrastructure has gained importance. Apart from large-scale map providers, also automotive manufacturers and traffic infrastructure providers rely on digital spatial information for different purposes. There are two main drivers for this field of research: (i) the need for precise spatial digital twins to enable simulation and driving functions, especially for autonomous mobility and (ii) the need for predictive maintenance and asset management of infrastructure. The range of modalities thereby spans from road over rail to naval mobility. This makes highly specialized mobile sensor systems necessary, which allow for large-scale data capture, high-precision geo-localization, automated 3D-modeling, segmentation and characterization of individual objects, as well as automated data and update management. Imaging sensors, be it cameras, LIDAR, radar or sonar, form the backbone of data capture and allow the application of computer-vision methods for pose estimation, large-scale 3D-modeling, object segmentation and damage detection. From a system perspective, diverse sensor data from profile and area-sensors needs to be synchronized and fused with precision geo-localization data. The resulting geo-referenced information needs to be cleaned, validated, stored and updated, which also calls for a high degree of automation.
The workshop aims at exchanging research results on mobile data capture, automated data processing and data management systems with the purpose of digitizing traffic infrastructure and assessing their condition. Underlying sensor technologies may include cameras, radar, sonar, lidar, as well as complimentary imaging systems. Data processing workflows and methodologies may focus on localization, data fusion, data cleaning, privacy and anonymization, 3D modeling, semantic understanding and defect detection.
Topics of interest
- Vision-supported geo-localization
- Mobile sensor systems
- Large scale 3D modeling
- Data cleaning and anonymization
- Semantic 3D scene understanding, object detection and segmentation
- Data management and update management
- Automated defect detection, classification and prediction
- Matthias Rüther, Joanneum Research
Workshop on Object Recognition and Manipulation (ORMR)
The ability to recognise and manipulate objects is central to robotics. For example, it might be useful for a robot to recognise a certain object requested by a user, and to determine if and how the object can be safely grasped in order to fetch it. However, despite decades of research, such abilities remain limited in practice. Some of the limiting factors are a paucity of usable, large data sets for training robust models for the tasks under study, a lack of objective evaluation protocols to test these models in a comparable way, and more generally the challenge of reproducing results.
This workshop will bring together the seven projects that have been sponsored under the CHIST-ERA project on “Object recognition and manipulation by robots: Data sharing and experiment reproducibility (ORMR)” to share their advances and create synergies. They invite an international audience to participate and discuss possible options to move from datasets to benchmarks on real robots.
Topics of interest
- Perceiving or predicting physical properties (shape, orientation, mass, fragility, etc.) of objects or environments
- Handling of unknown objects and environments
- Developing systems which are capable of operating in ambiguous contexts
- Managing the perception-action loop
- Interaction and cooperation with humans or other robots
- Designing safe, secure, robust and ethically-sound systems
- Independent and objective evaluation
- Criteria and measures for reproducibility
- Markus Vincze, TU Wien
- Krystian Mikolajczyk, Imperial College London
- Andrea Cavallaro, QUeen Mary University of London
- Mihai Andries, IMT Atlantique
- Berk Calli, Worcester Polytechnic