Visual Servoing at UNH, Day 3: Style points

I had written a new web-based GUI for the vehicle's onboard database, with some open flash chart graphing capabilities. When we recorded the overnight recharging of the Odyssey IV battery, my code got its chance to shine.

As you can see, cell 10 seems to be defective and it will probably be replaced soon.


Since we were able to closed-loop control our position and yaw based on the camera, and our altitude based on the DVL, it made sense to close the loop on sway based on our heading.

This means that we can say "look at the south side of the pipe", and it maintains a position on the south side of the pipe. In this nav plot, we visited the east, south, west, and north sides.


We packed up and headed back to the lab. Another very successful trip!

Visual Servoing at UNH, Day 2

The day started off very well, with us checking out the mounting pole for a piece of multibeam sonar equipment that was being tested earlier in the week.


Satisfied that the image processing component was working properly, we switched to the 13cm pipe that we would be using for our demo.


Almost immediately, we ran into trouble... all of our calculations for distance were off. We ran and re-ran our calculations for the angular resolution of the camera, and for calculating our distance from the pipe based on its width in pixels on the video.

An hour or 2 later, we noticed that we had accidentally declared our angular resolution as a boolean instead of a double. That's a pretty big error! Surprisingly, it only affected our distance calculation by a factor of ten. But once we fixed it, we were spot on.


The results are dramatic!

Visual Servoing at UNH, Day 1

Today we are at UNH to work on some visual servoing code that Justin wrote. The goal is to be able to follow a vertical pipe, controlling yaw and position based on image processing.


Mike has mounted the camera on the bottom of Odyssey, facing forward so that we will be able to track an object in front of the vehicle.


Early on, we had to re-adjust the aperature on the camera. We had made it as small as possible to increase our depth of field, but it was just too dark in the facility to see anything on that setting. So, we opened up the aperature and double checked the focus.


With the vehicle on an ethernet tether, we were able to use coriander to tweak the settings of the camera once it was underwater. This program makes the tuning stage go very quickly!

Night Testing: Camera Tests

At our second night at UNH, we planned ahead and bought plenty of food before the stores all closed. It helped keep us awake to take food breaks throughout the night, but we all took periodic naps when we could.


Although the camera driver and strobe system worked, the images were a little too bright (and slightly out of focus, even at long distances). We will have to adjust the focus and the aperture before doing any survey work.


At dawn, we packed up and left.

Night Testing at UNH

To tune up our new camera and flash system, we planned to take a trip up to UNH to use the large pool at their Ocean Engineering facility.

They were already booked for the week we wanted. But because we had a fairly urgent need to get our testing done, they agreed to let us use the pool "from 7pm until dawn". Time for some all-nighters! We loaded up the truck in the afternoon instead of the evening.


The UNH facility is more than large enough for our needs. It would be great to have access to a crane like this at a pier somewhere... or in our lab, or in an MIT pool facility.


It really makes testing more productive when you can get the vehicle in and out of the water quickly!

Our major accomplishment for the day (night) was getting the PID gains tuned for depth/altitude control. They are tight to within 1cm of the target altitude!


The one nice thing about working at night is that you can turn off the lights to see how effectively the strobe system illuminates the bottom.