Multi-Modal Learning and Sensing for Mobile Robotic Systems
Mario Fritz1 and Trevor Darrell
One of the most vital capabilities of mobile robotic platforms is to recognize and categorize entities in the environment. Tasks such as reasoning about the current state of the world, assessing consequences of possible actions, as well as planning future episodes build on such a basic "understanding" of what roles objects and places may possibly play. We are driven by the goal to enable such systems with this basic capability by making best possible use of the available sensing modalities. Therefore we plan to extend our research on visual categorization towards optimal sensor fusion and dealing with heterogeneous training data.
Figure 1: Example office desktop and telephone recognition result
1Postdoc, EECS & ICSI