"Java, Sun SPOT and the FIRST Robotics Competition"

[pogenwurst]

This just came up in Google Reader; after a quick search I couldn't find it posted anywhere here and thought it might be interesting to many CD'ers:

http://blogs.sun.com/roger/entry/java_sun_spot_and_the

[Branden Ghena]

That's very interesting. Java on FIRST robots... not bad. Its really good to get options out there to students, and the more programming languages able to run on FIRST robots the better it is for students.

Let the C++ vs Java rivalry begin.

[keehun]

This is not related to SUN but it's so welcoming to see some familiar pits in the article pictures!

[bobwrit]

Note to self: Review Java....

[virtuald]

I hate Java (its a religious thing though, no actual reason), but this is pretty cool. Now if they would just get rid of vxWorks and let us use Linux on there we'd be really set...

[AustinSchuh]

Originally Posted by virtuald

Now if they would just get rid of vxWorks and let us use Linux on there we'd be really set...

I run linux every day on my laptop and every other computer I own and use, but I wouldn't want to use it on a robot if I had the choice to use VxWorks or some other real realtime OS. The robot needs to be realtime to be guaranteed to respond quickly to user and sensor input, and Linux is not designed to do that. It's designed to run on servers and desktops where a bit of latency isn't an issue. There are patches that have been made to make the linux kernel more real time, but they aren't nearly as good as running an actual realtime OS.

[virtuald]

Originally Posted by AustinSchuh

The robot needs to be realtime to be guaranteed to respond quickly to user and sensor input, and Linux is not designed to do that.

So I don't know if you've noticed this or not, but how much realtime stuff did most people's code actually do this year? Answer: most probably did almost nothing. What about interrupts? Nothing again.

Thats what the FPGA is for, it takes care of the nasty latency dependent stuff for us (among other things).

It's designed to run on servers and desktops where a bit of latency isn't an issue. There are patches that have been made to make the linux kernel more real time, but they aren't nearly as good as running an actual realtime OS.

I won't disagree, but I don't think the realtime stuff is a huge issue for what most people are doing. If there were no FPGA, then I'd definitely agree. Otherwise... hard to say.

From the experience I've had so far, porting software to vxWorks is really annoying (even nix-like software) because it has random oddities in its networking library (and other libraries too). The biggest advantage I see for linux would be that theres a *ton* of software that would pretty much *just work* without too much modification. Not so with vxWorks.

Then we wouldn't be just talking about java bots, we could have python bots or ruby bots...

[AustinSchuh]

Originally Posted by virtuald

So I don't know if you've noticed this or not, but how much realtime stuff did most people's code actually do this year?

Correct me if I'm wrong, but I believe that updating the speeds of the motors on the robot 50 times a second while processing images using the rest of the CPU time would qualify. When I've tried to get code I've written under Linux to execute a loop 50 times a second without the timing being off by a fairly large percentage while using all of the available CPU resources, I've had trouble and found the best way to get the loop to execute every 20 ms is to use less CPU resources. Maybe I was doing it wrong, but that's been my experience.

[virtuald]

Originally Posted by AustinSchuh

Correct me if I'm wrong, but I believe that updating the speeds of the motors on the robot 50 times a second while processing images using the rest of the CPU time would qualify. When I've tried to get code I've written under Linux to execute a loop 50 times a second without the timing being off by a fairly large percentage while using all of the available CPU resources, I've had trouble and found the best way to get the loop to execute every 20 ms is to use less CPU resources. Maybe I was doing it wrong, but that's been my experience.

You're right, obviously linux and vxWorks have different design goals. Probably would be a bit better if you were running in the kernel like we do on the cRio or if it were a realtime patched kernel, but hard to say -- I've no experience with realtime Linux.

In any case, I value the flexibility quite a bit more than I value the realtimeness. Cmon, a robot controlled by a shell script... would be pretty amusing.

[bobwrit]

Originally Posted by virtuald

You're right, obviously linux and vxWorks have different design goals. Probably would be a bit better if you were running in the kernel like we do on the cRio or if it were a realtime patched kernel, but hard to say -- I've no experience with realtime Linux.

In any case, I value the flexibility quite a bit more than I value the realtimeness. Cmon, a robot controlled by a shell script... would be pretty amusing.

A)Programming a robot with a shell script is on the same level as programming it with a perl script, in my mind.
B) We were half hoping that someone, somewhere would port lolcode to vxWorks...

[virtuald]

Originally Posted by bobwrit

A)Programming a robot with a shell script is on the same level as programming it with a perl script, in my mind.

Hmm. If you could tolerate a small bit of latency, using my embedded webserver for the bot you could program it using javascript.

[lingomaniac88]

I think this move will attract a lot of people, especially in schools that offer computer science courses in Java. There are probably quite a number of them, too, since AP Computer Science works in Java.

I'm going to speak with my school's computer science teacher. Maybe she can get her students involved with the team also!

Maybe we can also get the College Board to work with FIRST to make a new case study revolved around the new library. Who's with me?

[pyr0b0y]

I think that Java working with devices is a great step...I personally do not like Java at all but it's nice to see how popular FIRST has become that you can even program in Java now. It might bring an increase to the amount of students that join robotics teams.

[davidalln]

I'm excited to dive into Java as an alternative after a subpar experience with LabVIEW this year. I do have a few questions about how much of the more advanced/third party Java solutions we can use that will still be executable on the cRIO. Unfortunately, I didn't get a chance to drop in and say hi to the people from Sun at the competition over the weekend.

For example, can we use Java-based wrappers and implementations for other languages such as JRuby (my guess is yes, because Java is compiled into bytecode anyways, so whatever third-party libraries used should be compiled with it)? And will there be the ability to use a testing framework such as JUnit (with the emulator, it might be possible. I'd just rather not have to build and load every time I need to test out a small tweak or new feature, especially in the process of developing said feature)?

[j_johnson]

Originally Posted by davidalln

I'm excited to dive into Java as an alternative after a subpar experience with LabVIEW this year. I do have a few questions about how much of the more advanced/third party Java solutions we can use that will still be executable on the cRIO. Unfortunately, I didn't get a chance to drop in and say hi to the people from Sun at the competition over the weekend.

For example, can we use Java-based wrappers and implementations for other languages such as JRuby (my guess is yes, because Java is compiled into bytecode anyways, so whatever third-party libraries used should be compiled with it)? And will there be the ability to use a testing framework such as JUnit (with the emulator, it might be possible. I'd just rather not have to build and load every time I need to test out a small tweak or new feature, especially in the process of developing said feature)?

I got a chance to talk to several of the people at the Java booth at Atlanta and learned quite a bit about the setup. It will use the Squawk JVM running on top of the vxWorks OS. The WPILib should be ported over to Java. (It looked like they had most if not all of it ported already) This VM is Java Eicro Edition (Java ME), which is what runs on phones, smart cards, etc. Any third party libraries would need to be compiled down to the Java ME byte code, limiting them to using only the Java ME API, rather than the Java Standard Edition API.

The build system uses nant, and they will have plugins for the NetBeans IDE and maybe the Eclipse IDE as well. (As a side note, it was probably not a good idea to ask the father of the Java language why they are using a Sun IDE rather than one developed by IBM.)

The cRio will basically be formatted for c++ with the initial program being a bootstrapper to start the JVM. It sounded like this initial JVM will then listen for communication with the development PC and start a second VM with the robot code in it. Hopefully this means that it will be much easier to reprogram and test without the 30 second start up time.

All of this, of course, was in pre beta and could very easily change before it is released.