Skip to content

Agile Projects

TECoSA Agile Projects provide a fast and lightweight funding mechanism for TECoSA researchers to test new ideas, explore promising use cases, and develop proof-of-concepts. It is designed for small, time-limited projects that can move quickly from idea to action. So far, TECoSA has funded the following projects through its agile project mechanism.

CORRIDOR – Perception Corridor in the TECoSA Outdoor Testbed

Contact: Gianfilippo Fornaro, fornaro@kth.se

CORRIDOR adds two roadside sensor nodes from XENSE Vision to TECoSA’s 5G outdoor testbed on the KTH campus. Together with the existing node, they form a “perception corridor” of three time-synchronised nodes. The nodes process data at the edge and share anonymised descriptions of road users instead of images. This supports traffic-flow estimation, collaborative perception and the extraction of traffic scenarios. The project delivers integration with DISCO, documented interfaces, a low-latency data service and an anonymised dataset covering all sites. A live demonstration of road-user tracking is planned for KTH’s 200-year anniversary in 2027, and the corridor will stay available to TECoSA researchers.


DISCO – Distributed Software Platform for Cooperative Autonomous Systems

Contact: György Dán, gyuri@kth.se

Autonomous drones and ground vehicles increasingly work in teams and need to share communication, computing and sensing resources. DISCO develops a software platform that manages these resources across many autonomous systems and shows what is available to higher-level autonomy functions. Tasks such as perception, planning and control can be placed across the device–edge–cloud continuum, linking robots, 5G radio access and edge computing. The result will be a modular prototype with clear interfaces to ROS2 and the TECoSA indoor testbed. It will make it easier for KTH researchers and TECoSA industry partners to prototype and demonstrate new Edge AI use cases.


E-MAP – Edge-based Perception and Mapping

Contact: José Manuel Gaspar Sánchez, jmgs@kth.se

To operate safely, autonomous systems must map their surroundings and track moving obstacles in real time. Transitional Grid Maps (TGM), a framework developed at KTH, does this well but currently needs desktop-class computers. In some applications, such as underwater robotics, the computation cannot be offloaded and has to run on low-power hardware on board, at the “far edge”. E-MAP will optimise TGM for predictable execution on edge AI hardware such as NVIDIA Jetson. It will also add uncertainty estimates, so the system can report its confidence to planning functions, and benchmark the trade-offs between map resolution, uncertainty and available computing power.


MARS – Multimodal Acquisition for Responder Safety

Contact: György Dán, gyuri@kth.se · Mattias Wildeman · Shraddha Tripathi, shrtri@kth.se

Lightweight, battery-powered devices such as smart glasses depend on offloading demanding tasks like mapping, tracking and object identification to the edge, and the data involved is often privacy-sensitive. MARS builds a research prototype of such a device that combines two tracking cameras, an inertial sensor (IMU) and a thermal camera. FPGA firmware handles synchronised data capture, timestamping and transmission. In the demonstration, the prototype is carried through an indoor space while a computer maps its path in real time and displays it together with the incoming sensor data. The platform will support future research on trustworthy, privacy-aware spatial processing at the edge, together with Ericsson Research.


SPRINT – Speculative Reasoning for Responsive Human-Robot Interaction

Contact: David Broman, dbro@kth.se · Zirong Chen · György Dán, gyuri@kth.se · Rafia Inam, raina@kth.se · Iolanda Leite, iolanda@kth.se · Diletta Olliaro, olliaro@kth.se · Merle Reimann, mmrei@kth.se

Conversational robots have to answer quickly to feel natural, but language models running at the edge must trade speed against answer quality. SPRINT develops an open, modular framework for faster AI responses at the edge. It compares methods such as speculative decoding, and an adaptive policy picks the most suitable one based on network conditions, available resources and the task. The framework will be built into a conversational robot and tested in a user study of response time, answer quality and how the interaction feels. Visitors will be able to talk to the robot and experience the difference. The project brings together three TECoSA postdocs with expertise in AI, human–robot interaction and performance analysis.


TrACe – Trustworthy CI/CD Supporting Continuous Assurance of Cyber-Physical Systems

Contact: Marcus Nolte, mnolte@kth.se

Autonomous cyber-physical systems are designed around assumptions about their environment and sensors, and they are only safe if those assumptions hold during operation. TrACe builds a prototype DevOps pipeline that automates the building, testing, deployment and runtime monitoring of ROS2 components, with test criteria derived from system requirements. Components are deployed through the DISCO middleware, with a focus on KTH’s AD-Eye automated driving platform in the TECoSA outdoor testbed. A dashboard will let developers follow and supervise the pipeline. The work lays a foundation for research on continuous assurance of safety-critical systems.