Laura Weihl will give a talk about her work on Probabilistic Property-Based Testing for Feature Classification in Submarine Vision. Details below.
Laura Weihl, PhD fellow, ITU.
Probabilistic Property-Based Testing for Feature Classification in Submarine Vision
Navigation and localisation for autonomous underwater vehicles (AUVs) is challenging. Camera input can complement the sensory data for the AUV navigation stack in small underwater environments and hence increase the precision of navigation underwater. To test the performance of visual simultaneous localisation and mapping algorithms (vSLAM) in the underwater setting we need new approaches to synthesise test data because traditional simulators fail to model the long tail of safety critical scenarios. In this project I propose new methods to generate underwater test data to make claims about the reliability of vSLAM. To this end, I evaluate components of the vSLAM pipeline on new and realistic test data synthesised by generative processes based on neural networks.