MILA: Middleware-Integration von Leitstellen- und Dispositionssystemen für Autonome Mobilität

Problem Formulation

The deployment of heterogeneous autonomous vehicle fleets (e.g., Level-4 shuttles from different manufacturers) in real-world environments introduces significant challenges related to uncertainty, system heterogeneity, and dynamic traffic conditions. A central issue is the presence of latency and information asymmetry in distributed fleet-management architectures. Decision-making relies on incomplete, delayed, or inconsistent information from vehicles, infrastructure, and middleware systems. At the same time, vehicles differ in capabilities, operational constraints, and reaction behaviors, making unified orchestration highly complex. Furthermore, dynamic traffic restrictions—such as temporary bottlenecks, road blockages, or degraded operating conditions—are not captured in static maps. These must be detected and interpreted in real time and transformed into standardized, machine-interpretable representations (e.g., constraints or events) suitable for optimization and decision-making. Finally, there is a lack of validated approaches for ensuring robust, interoperable, and real-time fleet coordination under realistic operating conditions, including communication delays, system faults, and multi-manufacturer environments. Overall, the problem is to enable robust real-time orchestration of heterogeneous autonomous fleets under uncertainty, while integrating dynamic traffic knowledge and ensuring practical deployability.

Solution Approach

To address these challenges, the project proposes a multi-layer, AI-driven orchestration framework combining optimization, learning, and real-world validation:

  • Predictive Multi-Agent Optimization: Development of real-time optimization algorithms that model fleet coordination as a multi-agent system under uncertainty. Vehicle-specific constraints, traffic conditions, and operational requirements are abstracted into standardized cost functions and constraints, enabling scalable and efficient decision-making.

  • Operational Situation Understanding (OSU): A transformer-based deep learning approach is used to interpret heterogeneous inputs (e.g., telemetry, infrastructure data, historical events) and detect dynamic traffic restrictions. These are translated into semantic, standardized events and constraints (e.g., “passable with delay”, “blocked for certain vehicle types”), including uncertainty estimates.

  • Middleware Integration via Abstraction Layers: A manufacturer-neutral Vehicle Abstraction Layer (VAL) ensures consistent fleet-state representation and seamless integration of heterogeneous vehicle systems into a unified orchestration pipeline.

  • Real-World Validation and Digital Twin Integration: The system is validated in real-world environments (e.g., Kaiserslautern test field and MoD operation scenarios) using a ROS2-based multi-OEM fleet emulator, enabling reproducible testing of interoperability, latency, and fault scenarios.

This integrated approach enables data-driven, real-time decision-making without interfering with vehicle-level trajectory planning, ensuring scalability and robustness.

Research Questions

In light of the above challenges, the project addresses the following key research questions:

  • How can heterogeneous autonomous vehicle fleets be orchestrated in real time under uncertainty, while accounting for incomplete and delayed information?

  • How can dynamic traffic restrictions be detected, semantically represented, and integrated into fleet-level decision-making processes?

  • How can multi-modal AI models (e.g., transformer-based approaches) be designed to provide interpretable and uncertainty-aware representations of operational traffic situations?

  • How can interoperability between different vehicle manufacturers and fleet-management systems be ensured through standardized abstraction layers?

  • How can the robustness, scalability, and real-time performance of such systems be validated in realistic operational environments?

Keywords

  • Autonomous Mobility

  • Middleware Integration

  • Interoperability

  • Fleet Orchestration

Funding

Time Span

August 2026 - March 2030

Project Partners

  • Leipziger Verkehrsbetriebe GmbH 
  • ADASTEC GmbH
  • Deutsches Zentrum für Luft- und Raumfahrt e.V.
  • DB Regio Bus Mitte GmbH
  • ebblo Germany GmbH
  • Eckardt Software Management ESM GmbH
  • Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V.
  • FZI Forschungszentrum Informatik
  • Hamburger Hochbahn AG
  • HOLON GmbH
  • Rheinisch-Westfälische Technische Hochschule Aachen
  • Institut für Klimaschutz, Energie und Mobilität - Recht, Ökonomie und Politik e.V.
  • Interlink GmbH
  • INIT Innovative Informatikanwendungen in Transport-, Verkehrs- und Leitsystemen GmbH
  • INYO Mobility GmbH
  • IVU Traffic Technologies AG
  • MENTZ GmbH
  • MOIA GmbH
  • MOTOR Ai GmbH
  • Münchner Verkehrs- und Tarifverbund GmbH
  • P3 digital services GmbH
  • Rhein-Main-Verkehrsverbund GmbH
  • Rheinland-Pfälzische Technische Universität Kaiserslautern-Landau
  • Stadt Leipzig
  • Technische Universität Dresden
  • Technische Universität München
  • ÜSTRA Hannoversche Verkehrsbetriebe AG
  • Verband deutscher Verkehrsunternehmen e.V.
  • Via Mobility DE GmbH
  • Verkehrsgesellschaft Ludwigslust-Parchim mbH

Contact

Prof. Dr.-Ing. Naim Bajcinca
Gottlieb-Daimler-Str. 42
67663, Kaiserslautern
+49 (0)631/205-3230
naim.bajcinca(at)rptu.de