KIGlas – Optimization of glass production with the help of artificial intelligence methods

The aim of the project is to reduce waste in glass production with the help of data-based forecasts. The research work focuses on automated production systems that process customer-specific orders using renewable energy. The methods used are deep neural networks for the predictive analysis of sensor data from the production environment. At the same time, the sensor technology is supplemented according to the needs of the analytics. Finally, a decision support system is implemented on the basis of the forecasts, which provides recommendations for optimizing and increasing the efficiency of production. This approach has the potential to stabilize glass production in Bavaria despite high energy prices and thus secure jobs in this industry.

Optimization of intermodal transport in rural areas – OptiModal

Problem definition

To create comprehensive and time-efficient public transport services in rural areas, demand-responsive services are increasingly being set up. As these services only function efficiently in very limited areas, they need to be linked to an adapted scheduled service. Planning such intermodal public transport as a competitive service to private transport is a complex planning and optimization problem that can hardly be solved without the help of suitable digital tools.

Project objective

The “OptiModal” project will develop suitable tools for the planning and optimization of such offers. The project is developing a simulation as a digital twin for the evaluation of offers. With the help of optimization tools, the offers can then be improved and re-evaluated with the simulation. Structural, movement and usage data from existing mobility services as well as analytical methods including neural networks will be used to develop the tools.

Implementation

The starting point of the project is the creation of a database as a virtual image of the region (data lake). This data is used to analyze transport requirements and the choice of means of transport. Based on this, a realistic simulation and optimization tools are implemented and interlinked. This allows planning steps, such as the positioning of stops and the creation of timetables, as well as their review and optimization, to be partially automated. For interaction with planners, the project is researching new visualization options for presenting the complex interrelationships in this planning context.

ROLAND – Rural Remote Operated Land Vehicle

Especially in rural areas, the supply of everyday goods is an ever-increasing challenge. Especially in rural areas with a low population density, it is already almost impossible to provide a cost-efficient supply of goods transportation. While various cost-saving concepts for local public transport, such as on-demand transport services or driverless shuttles, are being investigated or are already in use, there is little focus on innovative solutions for everyday goods. In particular, the delivery of goods to people with limited mobility is an important task for the future supply of goods in rural areas. Autonomous delivery systems can and must make a contribution here in the future.

Delivery robots face various challenges these days. In addition to regulatory and legal issues, there are also questions regarding acceptance and social aspects. However, probably the biggest challenge remains the complex automation of the systems. As in the field of highly automated driving, today’s manufacturers face major hurdles in terms of the safety and reliability requirements of sensor technology and environmental perception. In addition, aspects of theft and vandalism protection must be taken into account for delivery robots. These requirements limit the use of current delivery systems to urban areas, particularly outside Europe.

The aim of the project is to develop new concepts based on methods of (generative) artificial intelligence in order to use the limited communication infrastructure in rural areas for the realization of a reliable and robust teleoperated delivery system. Rural areas and their communication networks pose particular challenges to delivery systems, such as highly variable bandwidth availability. The project will deal in detail with the development of methods and solutions under the given constraints.

An important part of the project is the further development of existing direct teleoperation systems towards indirect (trajectory-based) teleoperation, particularly in areas with low transmission rates. Teleoperation has the great advantage that a large number of robots can be controlled from a control room, sometimes simultaneously. This means that the personnel required to deliver goods can already be significantly reduced without having to completely solve the extremely high challenges of autonomous driving. The flexible use of direct teleoperation, which is particularly dependent on low latency, and indirect teleoperation is intended to bridge limitations in the existing communication infrastructure.

Innovative concepts based on machine learning methods are used to reduce the latency and communication bandwidth requirements of teleoperation. In particular, data compression, reconstruction and prediction using generative models represent possible solutions for the existing limitations.

By using several robots and networking them via V2V concepts, as well as connecting them to a central server, advantages from the field of crowd-sourced data can ultimately be applied.

With the help of these concepts, a possibility for the realization of delivery robots is developed that is cost-efficient and safe on the one hand, but on the other hand could also be used commercially in the coming years. In addition, this approach enables the gradual further development of delivery robots towards greater autonomy. This can make a decisive contribution to the future supply of rural regions.