UP Board

[KurtEck]

Thought I would mention, that they have announced a new Kickstarter for UP2 boards...

https://www.kickstarter.com/projects...h-intel-apollo

Looks sort of interesting, but many of the improvements like multiple HDMI and Ethernet connections probably won't help on Robot...

Not sure if I will subscribe yet and if I do for what level...

[KurtEck]

Thought I would mention that I did break down yesterday to support the UP Squared Kickstarter. My guess is they will be several months late, but...

I went with the Celeron Ultra starter kit... I thought about going for one of the Pentiums, but figured the Celeron probably has plenty of horsepower for some of the robots.

Also as a side note: Also broke down and now have an RPI3 as I get the impression more people around here may be using them!

[jwatte]

My learnings have been that there are about three levels of compute power requirements:

1) Forward kinematics, peripheral control, basic data gathering: The Teensy and OpenCM 9.04 are great! The arduinos can even do it, although you can feel the limitations.

2) Real image/audio input, basic recognition (speech or image,) path planning and real-time SLAM: The Raspberry Pi 3 is certainly enough; direct camera input and GPU helps.

3) High-level AI, unassisted object recognition, real-time 3D point cloud derivation, various convolutional network applications: You need ALL THE POWAR!!! Jetson gives you a good GPU with some CPU to drive it. A single Intel CPU is not enough, no matter whether it's Celeron, Pentium, i7, or Xeon -- the GPU will run rings around it. However, even the bigger Jetsons are maybe not enough for the high-level AI-type work that you'd REALLY want to do in this area.

This means that the Intel CPUs live in a niche that isn't really that useful. I started out beilieving that it would be the right trade-off (several years ago) with mini-ITX and NUC form factor Intel-based computers, but they weigh too much, draw too much power, and under-deliver on the high-end modern AI-level computation.

[KurtEck]

Thanks Jwatte,

I totally agree with 1) especially with Teensy and the like.

But for 2) I think there are many valid options For example sitting on my table:


But choosing between boards can be interesting. Example in the above picture are 4 boards.

The bottom left is the RPI3.

The one to the right of it is the Odroid XU3-lite: would have put in the picture the XU4 instead (similar power to XU3), but with form factor including IO pins of the other three boards.

The one above the RPI3 is the Odroid C2:

And the one at the top right of it is the current up board.

All 4 boards have the same form factor, 40 pin connector, heat sink (or fan)... So for me not deciding factor. The Up squared however will be a bit bigger. The narrow side expanding make it more or less squared.

All 4 bourds want +5v power coming in. The RPI3 wants maybe 2.5amps, C2 similar, Up/XU4 maybe 3-4, again depending on what you plug into them.

I have not played enough yet with some of the newer boards. But I know from earlier boards like RPI2 and Odroid, that most/all of tue USB IO went through one channel on the processor and there was issues with IO throughput. With UP all of the USB connections have independent channels to processor... So if you do things like hook up primesense or other camera, plus USB based Servo driver, there were issues on getting higher level of frames or with HROS1 there would be driver pauses...

The main thing I was/am looking forward to, is if you are doing things like ROS where the primary development was on intel processors, you can simply install the main supported released. The ARM versions I tried at times had difficulties, like not all of the packages existed, some of them fault...

So I hoping it should just make life a little easier for things like that. But as I mentioned there are many valid options?

Anyway hopefully it will be fun!

[jwatte]

Oh, I'm sure you can build a fine robot with any of those boards!

My point is more that you can build THE SAME fine robot with any of those boards. None of them give you enough oomph to totally upgrade what you're doing in an algorithmic way.
Another way to put that same tier: Pick the board you like the most and that has the most software support for your use case, because while they each benchmark slightly differently, the end result is going to be in the same ballpark!

Another illustration: When your real-time point cloud reconstruction latency goes from 8 seconds to 4 seconds, it's a 100% speed upgrade! And it's still not useful for real-time navigation. Hence, why the totally different category 3 is needed.

[KurtEck]

Thanks again,

Sorry, I may not have been very clear in the last in the previous postings. I totally agree with you that none of these boards will take you up to your category 3. Which is again why I choose the Celeron version of UP2 as I think it will get me solidly into the higher end of your category 2 boards.

In the previous post, I was simply trying to show how there are now several options out there that have different architectures, that have roughly the same footprint and power requirements of the RPI... It was in regards to your statement about the earlier boards you tried like the NUC and ITX boards taking up too much room and power... I totally agree!

When Kevin first started his ROS robot, I also purchased a NUC, but never tried it on the PhantomX, as it was heavy/bulky and needed some external power up converter to get up to something like 17 or 21v... But as I mentioned after trying ROS on some of the different ARM based boards, thought it would be nice to be able to try out ROS again on an Intel 64 bit processor, where you hopefully can use the standard ROS repositories and hopefully be able to run all of the code on the main processor... Also hoping that having better USB hardware on these boards will allow us to get the full rated frame counts of the Primesense or equivalent camears and maybe resolve some of the USB latency issues we were seeing on RPI2s and the like.

I also totally agree, that you can build a fine robot with most any of these boards. Likewise I totally agree with you should choose the board that has the best support for what it is you want to do!

Merry Christmas!

[jwatte]

ROS being better supported on X64 is a great point, btw! While the embedded world is moving to ARM, the move doesn't feel like it's all complete yet :-)
Happy Christmas to you, too!

[KurtEck]

Thought I would give an update here...

The first batch of Kickstarter UP2 boards have shipped. They made a few changes to the board during the beta cycle, and we had a choice to have our kickstart fulfillment with one of the beta boards, or wait a month and get one of the updated boards. Also as part of the kickstarter stretch goals they are including a case, which was not ready yet... So again you could choose to wait until these were done or get a coupon to their store for the value of it... I decided to wait as the shipping cost would have been > then the coupon...
So I believe mine is scheduled to ship in July...

Some other updates: You can now order some of the UP boards from mouser: http://www.mouser.com/search/refine....&Ntt=167583189 Which should reduce the shipping costs

Also today they announced a new kickstarter: https://www.kickstarter.com/projects...ngle-board-com

Which is for a really small Quad core Intel Adam processor... For the heck of it may support them here again!

[KurtEck]

I thought I would mention, I did a big OOPS with my UP board.

I had my Teensy 3.6 HAT plugged into the UP board with the Pololu DC to DC converter, where I plug in a 12v wall wart (or lipo battery), which powers both boards... Well I plugged the 12v 10amp wall wart into the wrong connector, i.e. into the +5v connector of the Up board and heard a POP and smelled the smell you don't want!

So at least part of the board is toast (literally). I pulled off the heat sink and did not notice anything obvious. There are no schematics, so not sure if easy fix or not... I was told there is some power protection built in, so I may see if I can power around it and see if it will at least boot. Probably not worth paying to ship it back to Europe to see if it can be fixed and then if so ship it back...

And luckily their customer service people took pity and made me a deal, so will have a new one shipped to me soon...

But next time I will do, what I meant to do and tape over the power plug when I have the shield installed!

[jwatte]

I plugged the 12v 10amp wall wart into the wrong connector

Hate it when that happens! And I know exatly the smell you're talking about

My favorite power protection is a PTC resettable fuse followed by a high-power TVS followed by a MOSFET or Shottky diode.
If you plug it in wrong, the Shottky or MOSFET will protect against reverse polarity. And, the TVS will forward conduct, shorting the power, which will make the TVS resettable fuse trip.
If you plug it in right, but with too high voltage, the TVS will reverse conduct, and again, the TVS will likely trip (depending on how much current/over-voltage you use.)

The weaknesses of this design are:
1) The PTC resettable fuse has some trip time (often measured in seconds)
2) The TVS will conduct high-level current ("short") until the PTC trips, so it needs to be high-power (5kW) which means it needs to be big.
3) The PTC has some resistance, so will drop some power. If you use a Shottky, it will also have some resistance and drop some power (often up to half a Volt.)
4) If you use a MOSFET, you don't get the diode drop, but then have to worry about Vgs and avalanche (overvolting.) The TVS generally will help keep this in check, but see above.

So, all in all, the main problem is bulk and some power cost, to protect against a human error that "shouldn't" happen, and hook-ups only happen once, so the amortized benefit is low.

Anyway, good PR for the UP people to help you out; let's hear how it works when you get the new one :-)