Year's First Pool Test

The weekend 22nd and 23rd of January, we carried out our first pool test this year! This was an educational and engaging experience for us. We found many areas we can improve in, but there were also many things that worked as planned. The data we collected and the knowledge we gained from these tests will be useful as we continue to work towards our goals this semester.



The Autonomous Software group started at 8 in the morning on Saturday, with the goal of testing a virtual target following (VTF) system developed by a past Vortex member. This system allows the AUV to follow a path between arbitrary points in 3D. The system has worked very well in the simulator, but the same thing could not be said when Beluga entered the pool. Hours were spent mapping out the difference between the behaviour of the simulator vs. the real AUV, and the testing was eventually left for Sunday.



The Computer Vision and SLAM groups then arrived, and they as well came to the realization that much of which works very well above the water, does not work quite as well underwater. The colour detection systems struggled when the objects were not close by. Effects such as objects being mirrored underneath the water surface was also discovered, as well as some anomalies when performing camera calibration. But, great amounts of data was collected, and allows us to replay parts of the pool test in order to develop better detection and classification schemes.


On Sunday, the early autonomous birds again kicked things off, this time using the external camera system for localization, after determining that the localization debugging should be done off-site. Now with decent localization, the VTF controller proved to be very successful (after some debugging). The gates and markers were placed in the pool, and by hardcoding their locations, we managed to pass the gate! More data was collected in order to tune the simulator to match the real world. CV continued their battle from the office, now armed with lots of data to work with. However, some more data was collected and the pool test finally concluded.



In terms of hardware, most worked as expected - the only hiccup we had was with the killswitch and the camera connection. The killswitch can seemingly be attached, but without activating the magnetic switch on the inside. The camera seemed to have a bit of an issue with "low USB bandwidth". Overall though, Beluga performed very well. With the knowledge we have gained now, we are much better prepared for the prequalification.