Multimodal modelling and motion prediction in dynamic environments
Dr.-Ing. Birgit Graf
Fraunhofer-Institut für Produktionstechnik und Automatisierung IPA
- Universität Hamburg
- Fondazione Istituto Italiano Di Tecnologia
Competence Area: Embodiment
Tracking the precise location of humans and dynamic obstacles allows service robots to distinguish between places where they can operate safely and others, where collisions and thus safety-critical situations are likely to occur. Therefore, in this project, vision-based environment modelling, that is 3-D point cloud generation as well as geometric mapping will be integrated with other perception modalities such as sound-source localisation and touch sensors. Furthermore, motion tracking and prediction functions will be developed based on the integrated sensor input which will be provided in real-time to higher-level, cognitive modules.
Methods to efficiently process and store this enhanced environment map will be designed and implemented, that are specifically suitable for real-time collision avoidance. Besides providing functionality to give fast feedback on the risk level of planned actions, directly affecting behaviour by prohibiting dangerous actions, sophisticated real-time 3-D environment maps will also form the input for higher-level, cognitive safety modules, enabling them to make faster decisions based on more precise predictions.