We have been rather light with our updates in the previous few months but that hasn't stopped our progress. We have got a lot done and a lot to talk about so let's get started!
First off there's the Doppler Velocity Logger (DVL) from Teledyne which we've spoken so much about. You may recall our previous post talking about installing it to our frame, and after some technical hiccups ranging from simple to disastrous we produced the image below using real-time velocity information:
This image was made with a team-member pushing the robot around the edge of the tank, and after about 10 minutes, approximately 50 meters of swimming, and his estimation of where we started, the combination of the workhorse of our controls system, the Memsense IMUs, and the DVL had only accumulated about 1m of error! We're still testing the limits of the error of our system, but if that were the limit of our error, 1m could easily be corrected with visual information. And that was only the first try!
Since producing the above image we have integrated the software changes, which had been written months in advanced, and the controls system now properly uses the velocity information to perform tasks. We are continuously refining the estimation and movement systems, but they have quickly reached the stage that we are ready to use the new controls mechanisms in our AI. Check back soon to see us demonstrating these new controls in a cool and/or interesting way!
Another huge leap forward in Tortuga IV's design was the new cameras. Sidus Solutions has been mentioned several times in previous posts on the subject, and now we can unveil what they've helped us bring to life:
These new camera housings are, if you forgive a bit of self-indulgence, amazing. Designed by our mechanical lead Chris Carlsen, they are depth-rated beyond 300 meters, eliminate the previous design's heat dissipation issues, and are essentially the smallest housing that could hold both the cameras and our control electronics. We'd like to reiterate our huge thanks to Sidus Solutions for helping us build these housings! The electronics inside also deserve some special mention.
Here you can see another pair of our sponsors who helped make the new design possible: Advanced Circuits, who provided the printed circuit boards free of charge, as well as TC Technologies, who provided their incredible Eqcologics firewire chips free of charge. These boards are what allow us to pass what is normally a matched-impedance signal through the relatively messy interface that is both of our Subconn bulkhead connectors. All of this effort is meant to yield pictures for the robot vision software, so how about some samples? Now, these pictures may not look very impressive, but to those roboticists out there reading this: I think you can appreciate the quality of these pictures.
Traditionally, we had a lot of trouble differentiating between the yellow and green buoy, but the difference in these pictures is much sharper than with the previous cameras, and you can see that the noise in any given pixel is very low.
The new cameras also have a much higher depth of field than the previous cameras, which gives clearer images at a distance. You can also see that while the orange pipes were easily distinguishable before, they now practically pop out of the screen.
As in the first picture, normally we have trouble distinguishing between the different colors, but they are now vibrantly different, even in our vision software.
And as in the second picture, this demonstrates from nearly 40ft away that last year's targets are still easily visible, and distinguishable.
For comparison, here is a picture from our old cameras:
It may appear that the color definition on this camera looks comparable, but the grainy nature of this picture is actually a side-effect of the small pixel size (the new cameras have 7.4μm × 7.4μm for the cell size relative to the original 5.6μm × 5.6μm). That shifting noise made by the small pixel size plus the color bleeding and giving "soft" edges lowered the consistency of the previous system. You can also see the vignetting around the three corners of our camera view that our old housings created.
So we hope you appreciate this big update and we're planning on having a few more updates before the competition where we'll show you more of what our robot is capable of!
To end this progress update, here are two images of Tortuga IV for comparison: the final version rendered in our Pro/Engineer computer-aided design software, and the nearly-completed robot "in real life". We're almost there!