Autonomous Scale Cars

In the it:movES project, students collaborate with us to develop hardware and software for autonomous model vehicles on a 1:10 scale. Here, innovative solutions for autonomous mobility are being created, which have been and continue to be presented with great success in international student competitions, such as in the Carolo Cup (team it:movES) or the Bosch Future Mobility Challenge (team DriverlES). Our current software stack includes modern neural networks for image-based environmental perception, algorithms for sensor fusion, as well as action and path planning. An optimized NVIDIA Drive platform, in conjunction with ROS (Robot Operating System), serves as the hardware basis. This configuration closely aligns with current systems in the automotive industry, thus offering students a valuable opportunity to gain practical experience and optimally prepare them for future professional challenges in the field of autonomous mobility. Furthermore, the it:movES project promotes interdisciplinary collaboration and creativity by providing students from various fields the opportunity to work together on interesting issues in autonomous mobility.

Further information: 

• it:movES project webpage
• Press release "Erfolg für Team DriverlES" (2023) (in German)
• Press release "it:movES Team erreicht den 2. Platz beim Carolo Cup" (2022) (in German)
• Press release "Kleine Flitzer ganz groß" (2020) (in German)

Development of a Self-Learning Method for Improved Short-Term Wind Energy Forecasting (2022-2025)

Using spatially high-resolution, weather model-coupled CFD (Computational Fluid Dynamics) models based on the Reynolds-averaged Navier-Stokes equations (RANS), a high degree of accuracy in predicting wind speed has been achieved. This accuracy is crucial for assessing wind yields at potential sites. This success opens up the possibility of further developing these methods to create short-term forecasts for wind energy yields in complex terrains for wind turbines. The existing approach, which combines mesoscale weather models with artificial intelligence methods for forecasting (hybrid methods), will be enhanced. The focus will be on physical downscaling using detailed CFD simulations and the application of ensemble weather forecasts to determine the physical uncertainty in these forecasts.

Funded by the Deutsche Bundesstiftung Umwelt (DBU). 

Anonymized Collection and Use of Mobility and Movement Data (2022-2025)

For years, new forms of individual and collective mobility have been evolving. Service providers, including car-sharing and public transport, require insights from mobility data to tailor their services to user needs while conserving resources and protecting the environment. The "Anonymized Collection and Use of Mobility and Movement Data" (AnoMoB) project aims to develop anonymization techniques that allow providers to enhance mobility data collection and processing, while safeguarding citizen privacy.

Initially, the project will determine specific data needs of service providers, such as types of mobility data and attributes like age and income, required for planning resource-efficient, user-centric mobility. Based on these requirements, tailored technical solutions will be developed for privacy-compliant data collection, ensuring sensitive information is secured with robust protection measures. This approach also aims to boost user trust in service providers through reliable anonymization of their personal data.

Funded by the Bundesministerium für Bildung und Forschung

Predictive Human-Robot Interaction in a Collaborative Workspace (2020-2023)

In industrial settings, shared workspaces where robots and humans work together are becoming increasingly prevalent. Robots excel in executing simple tasks with precision and speed. However humans are often still required for complex tasks, leaving robots as underutilized and costly resources in these shared spaces.

This project aims to enhance work efficiency by exploring human-robot collaborations. A key objective is to develop methods for predicting human movements and actions within the robot's workspace. With these predictions, the robot's motion planning can be adapted accordingly, optimizing the collaborative process and resource utilization.

Publications

D. Lagamtzis, F. Schmidt, J. R. Seyler, T. Dang, and S. Schober. Exploiting Spatio-temporal Human-object Relations Using Graph Neural Networks for Human Action Recognition and 3D Motion Forecasting. International Conference on Intelligent Robots and Systems (IROS), 2023. https://doi.org/10.1109/IROS55552.2023.10342491

D. Lagamtzis, F. Schmidt, J. R. Seyler, T. Dang, and S. Schober. Graph Neural Networks for Joint Action Recognition, Prediction, and Motion Forecasting for Industrial Human-Robot Collaboration. International Symposium on Robotics (ISR), 2023. https://www.vde-verlag.de/proceedings-en/456140005.html

D. Lagamtzis, F. Schmidt, J. R. Seyler, and T. Dang. CoAx: Collaborative Action Dataset for Human Motion Forecasting in an Industrial Workspace. Proceedings of the 14th International Conference on Agents and Artificial Intelligence - Volume 3: ICAART, pp. 98-105, 2022. https://doi.org/10.5220/0010775600003116.

Autonomous Robots in Unstructured Environments (2022-2024)

The Mobile Robotics research project is dedicated to exploring innovative methods enabling mobile robots to operate autonomously in unknown, unstructured environments. Emphasizing sensor-based perception and self-localization, the project leverages cutting-edge machine learning techniques.

In its initial phase, the project involved a thorough analysis and evaluation of existing methods, primarily focusing on robotics in known and structured contexts, such as indoor logistics in industrial settings or autonomous road traffic navigation. These methods, however, proved to be less applicable to uncharted and unstructured environments.

The current project phase delves into adapting these existing methods for more complex, unstructured environments. This involves a comprehensive derivation of new requirements and the creation of a sophisticated system for evaluation and comparison. This system will not only encompass theoretical analyses but also include experimental studies using publicly available research datasets.

AMEISE (Phase III) - Holistic Research Into the Potential of Autonomous Driving in Public Transport (2023-2024)

In the fall of 2023, the emission-free and autonomous shuttle AMEISE progressed into its third phase of funding. This phase continues to explore how the shuttle can be seamlessly integrated with urban development and engage all relevant stakeholders. The focus is on identifying key factors that will drive the success of this initiative, leveraging expertise to optimize the shuttle's role in the urban landscape.

Further information: AMEISE webpage

AI-Based Quality Prediction of Injection Molded Parts (2022-2024)

This project's goal is to develop an AI-based forecasting system for predicting the quality of parts in injection molding.

Unlike current technology, this system is designed to significantly surpass existing prediction methods that rely on process monitoring through machine parameters. Furthermore, it aims to require significantly fewer training data than conventional AI applications, making it attractive for small-batch production.

The prediction model developed in this research project is intended to monitor the parts with such accuracy during the production process that post-production quality control becomes redundant.

To achieve this, appropriate sensors will first be integrated into the process. A multi-stage evaluation system allows for rapid deployment of the system, as it requires only a small amount of training data. The multi-stage classification enables the identification of unknown process states and their inclusion during ongoing production. Known process states can be robustly and securely classified by the system using machine learning methods.

Funded by the Ministerium für Wirtschaft, Arbeit und Tourismus Baden-Württemberg.