About
Invariant trajectory descriptors are trajectory representations that have been made independent of reference frame (or viewpoint), reference points on the object, motion profile and/or scale. Invariant descriptors are defined using concepts from differential geometry, Lie group theory and screw theory.
Invariant descriptors are used to construct coordinate-invariant trajectory models for different tasks. These trajectory models can be used to robustly recognize human motions in different contexts or to execute a human-demonstrated task on a robot. Thanks to the invariant properties, the models can be more easily generalized and extrapolated to different contexts compared to models based on trajectory coordinates that were recorded in a predefined reference frame. As a result, fewer demonstrations are required from the user to build generally valid models.
Our software offers the following features:
- Definition and robust calculation of new invariant descriptors
- Application of invariant descriptors to recognition and segmentation
- Offline and online adaptation of invariant trajectory models to novel contexts
Acknowledgments
The development of these software packages has been supported by the following projects and organization:
- ERC Advanced Grant 788298 (Robotgenskill) of the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme.
- FWO project G0B6613 of the Research Foundation Flanders (FWO).