INESC TEC PhD student Bernardo Teixeira did a short-term internship at
Edinburgh's Heriot-Watt University under the guidance of Prof. Sen Wang.
For 3 weeks, Bernardo studied
novel training and refinement techniques for motion
estimation within deep learning - based visual-acoustic pipelines. In
addition, alternative methodologies and new ideas for leveraging
acoustic data in the context of robotic motion estimation were explored
and discussed, namely what concerns to sensor fusion and regressing a
statistically relevant measure of uncertainty.
The resulting rosbag file containing synchronous camera image, acoustic
multibeam point clouds and groundtruth trajectory data from the UX1 Neo robot
is available for download in the following link: http://lsa.isep.ipp.pt/~bteixeira/ . With
this data, it is possible to benchmark the performance of the visual-acoustic
odometry algorithms being developed in an underwater scenario (i.e. CRAS lab
pool)