Ever wondered what makes a limelight 2 tick?

So did I, so I took one to bits.

Photo album: https://photos.app.goo.gl/Ld5mwoHty3imsdpFA

Outside

The case appears to be injection molded ABS. There’s an Ethernet port, USB port, Weidmuller terminal block, Micro USB, and a fan. The case has some ventilation slots. Visible from the top are 8 SMD LEDs with focusing cones. On either side, the center two LEDs are slightly offset to the side. They look like these Cree XP-E2 LEDs. There’s a fairly large camera.

Inside

The case is held together with four T6 screws. The case unfolds with just a ribbon cable connecting the mainboard to the camera. There are three PCBs: the camera module, the LED board (labeled "Lime Light LED Board v3.2.2 20191104), and the mainboard (labeled "Limelight Mainboard rev3.20191104 B1).

Camera Module

No branding identifies the camera module, but it appears almost identical to this sainsmart one, except for a slightly different lens. This would mean an OV5647 sensor. This checks out with the limelight FPS specs, and the fact that the only officially supported CSI sensors are the OmniVision OV5647 and Sony IMX219.

LED Board

Four two-pin headers connect the LED board to the mainboard. Again, I’m not exactly sure, but these LEDs look just like Cree XP-E2’s. Pairs of LEDs are driven by four identical LED driver circuits. The Diodes AL8860 LED driver chip is used. Also on this board are the status LEDs.

Mainboard

Taking off the LED board we can get a clear view of the mainboard. On the top are the four two pin sockets for connecting the LED board, as well as all the peripheral ports, and a connector for the 20mm fan. After unscrewing two loctite-d screws we can flip the mainboard out and the first thing we see is the brains: a Raspberry Pi Compute Module 3+. A DDR2 SODIMM connector like this one is used. Two more T6 screws are used to secure the compute module down. Pop the sides out, and the compute module folds up and out, exposing the rest of the board. I’m not experienced enough to say too much about the mainboard, but I can spot an Ethernet controller and some voltage regulation stuff. I’m guessing there’s not much to see here, just what’s necessary to power an RPI compute module off ~12v, use the camera interface, Ethernet, and USB, as well as some PoE stuff.

59 Likes

Stuff used:

  • ENC28J60 ethernet controller
  • FSUSB42 usb switch
  • TPS62143 5v step-down converter
  • AP3429 3.3v and 1.8v step-down converters (interesting that they switched from LDOs on the older models)
  • AL8860 LED drivers
  • J0006D01BNL ethernet port with integrated magnetics (this is a good ethernet port because it has integrated magnetics and allows you to use passive PoE)

And FYI I just added a bunch of images to the google photos album.

2 Likes

I hope nobody creates a vision tracking module that coincidentally has the same pinout as the limelight 2+ and would therefore be compatible with the limelight image. Just to make sure nobody makes that mistake, here’s the limelight 2+ pinout:

  • CAM1 LED: GPIO30
  • CAM1 Shutdown: GPIO3
  • CAM1 SDA: GPIO28
  • CAM1 SCL: GPIO29
  • Camera is connected to CAM1 pins (the device tree is configured for two cameras? but CAM0 never gets detected of course) correction: logical CAMERA_1 in the device tree is setup to use the CAM0 unicam pins. the camera is connected to the CAM0 pins
  • Normal ENC28J60 overlay pins - SPI pins CE0, MISO, MOSI and SCLK (GPIO8-11 respectively), as well as GPIO25 for a falling edge interrupt
  • Ethernet Clock: GPIO44
  • LAN Reset: GPIO31 (however, I don’t think the LL has this connected, just a pull-up to 3V3, which is fine, it just means the ethernet controller won’t reset when the pi does)
  • GPIO4 controls D9 (status LED closest to the edge)
  • GPIO5 controls D10 (innermost status LED)
  • GPIO13 PWM output sets CTRL for U23 and U24 (right LED cluster)
  • GPIO18 PWM output sets CTRL for U21 and U22 (left LED cluster)

15 Likes

I wonder if they’re trying to lower the power draw so it can be powered by the VRM rather than directly from the PDP. A Raspberry Pi with some accessories really shouldn’t draw more than 10W, so it’s kind of ridiculous that the Limelight takes up its own dedicated PDP slot with a 5A breaker.

The hardware has definitely seen some improvements since the Limelight 1. Does it look like they added more protection on the power inputs? I know with the Limelight 1, all it took was one short or reverse application of power to fry the internals (which we learned from experience). For a $400 piece of hardware, that’s really fragile. It always felt like a $50 piece of hardware with $350 worth of software but it seems like they’re making big steps in the right direction with the hardware.

The teardown and component identification was interesting and kind of cool. And if you wanted to build one yourself based on this information, it sounds like a fun project.
But… encouraging somebody to create a competing product by just stealing the best software around is kinda gross.
@lucas.kinion made the point that most of what you’re paying for is the software. And he’s right. And a whole lot of teams were happy to pay for it this year. It’s pretty great.

19 Likes

Other than price(!) the main gripe about the limelight is the bright LEDs. (A separate thread could probably be created to describe just how bright they are!)

If the visible green LEDs were replaced with IR LEDs, would the resulting system work the same and be far less objectionable? If so, this should move the radiation band beyond the visible spectrum.

Some imaging arrays do not detect the near IR band.
If there is enough energy - even in the IR band - it can still stimulate visible band receptors, such as eyes!

To be completely honest, at this point, I don’t know why you would copy limelight. Chameleon Vision seems to have gotten to a point where it is keeping up with the software side of Limelight, and I’m only expecting it to get much better. If you slap that on a Jetson Nano, get some LEDs, a camera, and if you really want to be fancy, a 3D printed case, that’s a vision processor that will be at least as fast as Limelight and have some nice software to go with it(for less than half the price).

7 Likes

No, thanks.

9 Likes

Not even a Jetson Nano. The devs recommend a Pi 4 and a PS3 camera and they claim it can track at over 100 FPS.

Well, I was saying if you are ambitious like myself and might want to look into PnP(which needs high corner accuracy), but of course you could always use one of those newfangled Pi4s

I respectfully disagree completely. Although the Limelight featureset isn’t trivial, it’s also not something that a few people can’t put together in their free time (see: Chameleon.) Pretty much every flagship Limelight feature exists in Chameleon vision, which runs at similar framerates on comparable hardware anyway (and Chameleon has other features like camera calibration that don’t exist in the Limelight but are very important for accurate PnP.)

What you’re actually paying for is convenience and system integration. The only reason not to use Chameleon at this point is because you don’t (or can’t) plug a camera into a Raspberry Pi, flash an SD card, and wire an LED ring. If you think you can do those things then you might start thinking about why you’re paying $400 dollars for someone else to do them for you—I am not saying this is an invalid choice, however. Some teams don’t have enough humans to do this or don’t want to spend the time.

6 Likes

We used Chameleon on a Pi 4 this season and it worked well for us. Still a lot to learn about OpenCV and PNP, but the built in models were good enough.

With this break in events I would love to see people contribute back to the project to make it even easier to use. We ended up attaching some LEDs to the GPIO ports of the Pi to reassure ourselves that it had started and was working.

3 Likes

I don’t recall telling anyone to create a competing product by stealing “the best software around”. All I’m saying is if someone were to create a raspberry pi based vision module with the same pinout, it would be compatible with their software. Really there’s only a few ways to use a CM3+ in this application, so it’s actually pretty likely that someone would coincidentally pick the same pinout. And btw, if someone was actually interested in getting this pinout info to make a competing product, they wouldn’t need this post to figure it out. It’s very easy to look at the device tree and follow some traces. It was mostly a joke.

The limelight is a glorified raspberry pi and LED ring. It’s design is not anything spectacular. Take a look at the CMIO or RPi schematics. Most EE students could make a limelight. It has some neat software (but ChameleonVision is good too) and they’ve learned some lessons that add value to their product. Stop worshiping them and drinking the kool-aid. They’re a company. They don’t care about FRC teams, they care about money. And with no competition, they’re free to charge an asinine amount of money for their product, to the point where it’s inaccessible to many teams.

5 Likes

I don’t really know what goes on in the minds of the Limelight company. What I can report is that we developed our own Axis camera-based vision system in 2016, and a Jetson based system in 2018 (which was never fielded). Despite those options, we bought two limelight’s since then. It has been one of the rare products that completely satisfied my expectations. $500 bought us a vision system that just works.

If someone uses this thread as inspiration to make a competitor that works as well - that’s great, please do! Lord knows FRC teams could use to save some money. In the meantime though, please stop throwing shade on the company. It is clearly not over priced if my team and hundreds more have decided the product is worth spending the money.

9 Likes

$500 bought us a vision system that just works.

That’s cool. I’m happy for you. Most teams can’t drop $500 on vision.

It is clearly not over priced if my team and hundreds more have decided the product is worth spending the money.

Not gonna argue over the definition of overpriced, but this is mostly due to the lack of competition. You don’t really have another option for a comparable vision camera right now.

1 Like

Do you think another comparable option should exist? Great, so do I. So what’s stopping you? Otherwise what argument are you trying to make?

1 Like

My definition of overpriced involves profit margins. Yours involves what’s available. I’m saying the limelight is overpriced (has crazy high profit margins), so someone needs to release a competing product. I was disagreeing with your statement that the limelight isn’t overpriced simply because people buy it.

You realize Limelight was started by someone in FRC right? Why do you get to decide the value their product is worth? Nobody is forcing anyone to buy it.

Saying they don’t care about teams when they delivered the first true easy to use out of the box vision system tailored for FRC is a bold strategy. Let’s see how it plays out for you in this thread.

25 Likes

Where are you getting this information from? Anytime myself or other teams I know of have had issues, Limelight has provided stellar customer service. This is not what I would expect from a company that doesn’t care about FRC/their customers.

When they first released a number of years ago, there was absolutely nothing on the market that came close to offering what they did. Limelight took vision (at the very least in my region) from something that only the most elite of teams could reliably perform, to something that most mid tier/even some rookie teams could accomplish. The way I see it, Limelight has done a massive service to the FRC community, by not only providing this product at what I would say is not an UN reasonable price (sure, I’d like everything to be cheaper), but also creating inspiration for many teams to develop alternatives (ChameleonVision included).

Where did you find the profit margins of the limelight published? I’d be very interested in having a look at this.

2 Likes

Crazy high profit margins? How do you know? How many do you think they sold?
This is a very small market, you can’t spread development, manufacturing and support cost over a lot of units. Also re-sellers need their cuts too. Make 10K of them in China then sure it would be overpriced.

For a team it’s a matter of resources. Spend $400 and my programming team has an instant solution, sign us up instantly.

7 Likes