The AVL aims to create a cooperative localization framework through which a team of robots can share information (eg. range, bearing, pseudo-range etc.) among themselves to allow them to update their positions to be consistent with one another. While similar work as been done in this field, we are looking to develop techniques where the data shared between agents are limited and where communication is more sparse.
The AVL dedicates a significant portion of its resources to developing new state-of-the-art state estimation methods. Among the areas of interest, the AVL’s main aim is to develop online robust estimators for hardware implementations. Effectively, the AVL seeks to develop estimators with a reasonably high level of robustness to nonlinearities and uncertainties while maintaining algorithms complexity level at a minimum.
As a lab, our work on machine learning is focused primarily on understanding the constituent parts of neural networks, individually and in combination, in order to extract the maximum performance with limited data and limited computational research. This is especially important for autonomous navigation in novel environments. A core component of this effort is identifying the unstated assumptions that are built in to common neural network architectures which can affect performance.