AI-FUSE-ROBOT: AI For Universal Scene Understanding: Robotics (SAIA202500X163851SV0)
AI-FUSE is a research project in the field of artificial intelligence focused on scene understanding, with the goal of developing AI capable of explaining what happens in images and videos in a way similar to how a human would. Coordinated by the University of Zaragoza, with the collaboration of Cima Universidad de Navarra, AI-FUSE aims to develop general scene understanding algorithms and AI-based tools able to analyse images and videos and explain the content of any scene, without being restricted to a single domain. The consortium will validate these tools in three key areas: robotics, cell biology and environmental monitoring, using real data provided by strategic collaborators. AI-FUSE was one of the twenty consortia selected, out of more than two hundred proposals, in the State Research Agency call for research projects in artificial intelligence. Within this framework, the Service Robotics Lab leads the AI-FUSE-ROBOT subproject, which addresses a key challenge in robotics: enabling robots to operate autonomously and robustly in environments shared with people. Current robots often struggle in crowded and dynamic scenes, while AI models face the limitations of onboard computing power and the lack of training data. The goal of the subproject is to build robots that move easily among people and interact with them naturally, which requires a much more human-like understanding of the environment. This technology will be key for the next generation of service robots in areas such as hospital assistance or smart logistics. To this end, AI-FUSE-ROBOT introduces three key innovations:
- Advanced sensor fusion, combining computer vision with LiDAR and other sensors to detect people and to track and predict their motion.
- Integration of real data, simulation and self-supervised learning to reduce the need for labelled data.
- Efficient AI models optimised to run directly on board the robot, enabling real-time responses without relying on the cloud.
In addition, the project will develop new tools and benchmark datasets to support progress in these techniques.
- Running period: 2025-2028
- Funding agency: MICIU/AEI/10.13039/501100011033