Skip to main content Skip to main navigation

Publikation

Amenable — Accessible AI in Sensor-Based Systems for Environment Monitoring

Daniel Sonntag; Martin Atzmueller; Frederic Theodor Stahl; Thiago Gouvea; Martin Günther; Christoph Manss; Lisa Dawel; Siting Liang; Felix Igelbrink
German Research Center for AI, DFKI Technical Report, 2026.

Zusammenfassung

Amenable is a three-year DFKI research project (2026–2029), supported by approximately 5 million euros in external funding from BMFTR, that advances Accessible AI for sensor-based systems in environment monitoring. The project unites three DFKI research departments—Interactive Machine Learning (IML), Cooperative and Autonomous Systems (CAS), and Marine Perception (MAP)—around a shared methodology: integrating knowledge graph and ontology engineering (KGOE) with robotics, interactive machine learning, and natural language inference. The central challenge addressed is the gap between the ad hoc, task-specific concept hierarchies common in robotics and sensor systems, and the formal, reusable knowledge structures developed in ontology engineering. Amenable bridges this gap through three interlocking components: semantic scene understanding via 3D Semantic Scene Graphs (3DSSGs) built and fused across multiple robotic agents; ontology-informed interactive representation learning that populates a knowledge graph from sensor data via expert-in-the-loop methods; and explainable, resource-efficient edge processing that embeds interpretability and energy efficiency into smart sensor systems. Natural language and visual entailment are employed to propose, assess, and rank candidate assertions, flag potential inconsistencies, and support consistency maintenance, while formal ontology reasoning and constraint validation provide the authoritative consistency layer. The approach is validated in a primary use case of indoor building exploration by a team of heterogeneous robots and assessed for transfer to ecological monitoring. This report describes the scientific goals and research questions, the project structure, and the expected contributions of the project.

Projekte