GEIHSR develops a generalist edge-intelligence framework for safe, low-cost, human-centered industrial robotics. The aim is to close the gap between perception and action by combining AI-based perception and certified safety with edge-optimized Robot Foundation Models (RFM).The approach rests on two pillars: (1) Robot Learning with RFMs and "Harmonic Mobile Manipulation", treating navigation and manipulation as a unified control problem; (2) Human/Spatial Reasoning with "System-2" reasoning and dense mapping (X-SLAM) to predict human intentions and move safely in dynamic environments.DFKI is responsible for semantic scene understanding, human intention prediction, and a standards-compliant safety stack with dynamic safety zones. Applied methods and tools include vision-language models, 3D sensor fusion, continual learning, and digital twins. A Real-to-Sim-to-Real pipeline bridges the sim-to-real gap.The system is validated in an industrial demonstrator in DFKI's human-robot collaboration lab. The project raises the maturity level from TRL 4 to TRL 6 and democratizes advanced automation for SMEs.
Partners
- EngRoTec GmbH
- Solectrix GmbH
- National Taiwan University
