So, I should start out by saying that I own a TK1. I do not own an OpenMV camera.
However, as much as I love my tk1 (and I really do!), I think that the OpenMV camera would be a better item to put in the kit of parts, rather than the jetson products. To make the jetson work with the robot system, I had to use 0MQ and c++ and get a dumb ethernet switch. With this we could just hook it up on analog or mxp and throw some values over the pins. I guess you could do this with the jetson GPIO but it would be evil and unreliable.
What are your thoughts? Take a look at the OpenMV platform (http://openmv.io) and maybe we can get the attention of someone at first.
-Gus
———————————————
FRC 3135 - Programming
FTC 3507 - Programming
LONG LIVE HASKELL
So you want to replace the Jetson TK1, a full SoC featuring a NVIDIA Kepler GPU with 192 CUDA cores, a 2.32GHz ARM quad-core Cortex-A15 CPU, 2GB of DDR3 RAM, 16GB eMMC besides countless peripherals all for $192 market price for an OpenMV cam with a 180MHz processor, 2 MB flash memory, 256 kB RAM and the limitation of only programming with Micro Python at a $70 market price?
I mean I get where you’re coming from with the concern on setting up such a complex system on the Jetson, but in my 4 seasons of competition, I’ve never felt that FIRST has had the need to simplify any aspect of the competition. After the incredible projects we’ve seen released by the community in the last years (Cheesy Vision and their path planner, the 971 state machines, this year’s OpenCV vision code for a driver station and many more), I believe that the best path for students and mentors involved in the control system for their teams should be adopting the greatest and latest in tech from the industry. Students in general have shown their fierce interest in using these technologies and I think that limiting their exposure to fields such as computer vision or embedded computing with these products, which implement tech such as Micro Python for getting to the rookie users in the market, will severely limit the learning possibilities for them.
as the person who wrote Tower tracker, the program I think your talking about, The biggest frustration that I have with first in my 4 years it is that programming needs to be rewarded…like this year driving forward fast for 15 seconds is enough to get you 10 points! and driving too far forward is not penalized but could kind of be rewarding because it could cause problems for the other team. Driving into the player stations at full speed really should have been a penalty if intentional or not. I understand hook and loop tape should be used but if a team doesnt know how to stop a robot from going too far they should not be getting points for their auto…HOWEVER programming is rewarded in the fact that generally the more well programmed a robot their is the better the robot functions but that does not mean you can have a blocker bot turn into a 2 ball auto bot…
OpenMV and the Jetson are not even remotely comparable. For one thing, the Jetson TK1 is actually available. Last I heard, OpenMV was struggling with manufacturer problems with bad solderballs on one of their chips, never mind shipping product out to their initial investors (kickstarter). It’ll be a while yet before they even could consider shipping product on the scale they’d need to to be in a kit of parts.
On top of that, OpenMV and Jetson serve different purposes. OpenMV is specialized for machine vision, as the name would indicate. I’ve seen some demos of it, and it’s actually pretty impressive what they can do with so little processing power, but you can’t do anything else with it. The Jetson is pretty much a full computer in terms of processing power and you can have it do a lot more. I know some universities have gotten 20 or 30 of the things and made small clusters of them.
Not necessarily. That’s why I made the poll. I think that having both would be nice. Also, the openmv platform outperforms the jetson is certain applications. It is all about knowing what hardware to use.
I agree completely. However, using the right hardware is a useful thing to know. While I can and have written CV code in java and c++ on a jetson, using a piece of dedicated hardware is more elegant, and I think it shows that you know how to be efficient and effective.