Automated vehicles, delivery robots and AI-based methods open up new opportunities for freight logistics on the last and second-to-last mile. However, whether such solutions prove their worth in everyday urban life depends on how they interact with the entire logistics network and the people on site. New approaches therefore not only have to work technically, but also need to be tested under real-world conditions. For start-ups and small and medium-sized enterprises (SMEs) in particular, setting up their own test environments is a major hurdle. The goal of the Last Mile City Lab (LMCL) project is to establish technology-neutral test environments in which science and industry can trial new services, products and solutions at an early stage of development.

The LMCL builds on the results of its predecessor project efeuCampus, in which a living lab with autonomous delivery robots was established in Bruchsal. The focus is on technology-neutral building blocks that work together in a modular way: real and virtual test environments, vehicle and sensor platforms, open AI methods and data, software solutions and operational concepts, as well as transfer formats. Open interfaces are intended to enable third parties to use these building blocks and extend them with their own developments. This is meant to lower barriers to entry and accelerate the development of new solutions for urban freight logistics in the region.

The IFL contributes to this goal with an open dataset and an associated virtual test field for autonomous delivery robots. Suitable datasets and simulation environments for training AI models and validating new methods on a comparable basis are still lacking in this domain. Central to this work is the creation of a multimodal dataset for autonomous systems in urban environments, in particular for sidewalk-based last-mile systems. A modular sensor platform forms the basis for this. Among other things, the dataset captures temporal and seasonal changes, which are rarely considered in existing datasets. Once processed and annotated, it is to be published as a freely accessible benchmark for semantic scene understanding. The virtual test field represents the real environment as a digital twin and is closely linked to it through a bidirectional data flow. Synthetically generated data thus support the training of the models, and new methods can first be validated in simulation before being tested in the real world. Together with open-source software components, this is intended to create an environment for training, validation and testing in which research and industry can compare methods for autonomous delivery robots under realistic conditions.

In addition, the IFL works on autonomy functions for delivery robots and their integration into heterogeneous fleets. This includes methods for semantic scene understanding, robust navigation without high-definition maps and the transfer of concepts from intralogistics, such as VDA 5050, to the urban environment. The methods developed are to be tested and demonstrated with various mobile robot platforms in real-world settings.

The project is funded by the European Regional Development Fund (ERDF) and the State of Baden-Württemberg. The funding period runs from 9 September 2025 to 31 December 2027.