Programmers: I Have A Challenge For You

[davidthefat]

I am bumping this one up cause I honestly did not want to start another thread and as the kickoff gets closer, I thought people should see this.

I found a solution to the image processing problem.
http://microcontrollershop.com/produ...ucts_id=352 9

No battery: Legal
4 USB ports = Good
Runs Linux (The Arm 9 Linux uses 32 mb of ram, and this has 32 mb ram... So IDK...)


I realized I can not use the PS3 or a PC due to lack of voltage generated form the batteries... Don't they need a minimum of 110v? and the battery is 12 v.

It says it supports up to 4 webcams, but I do not know how well it can perform while processing images.
I can either just process the data and send the raw data obtained from the camera to the cRio and process the logic there or process the logic on the board. Thats a minor detail that can be experimented with.

[Chris is me]

Originally Posted by davidthefat

I realized I can not use the PS3 or a PC due to lack of voltage generated form the batteries... Don't they need a minimum of 110v? and the battery is 12 v.

That's not exactly true. They need 110V of AC power, and the battery supplies 12 V of DC power. It's not really a matter of simply "oh, 110 is bigger than 12, guess I'm screwed". You could use a power inverter, though I don't think power inverters are FIRST legal...

Have you written any successful programming logic since your last post?

[davidthefat]

Originally Posted by Chris is me

That's not exactly true. They need 110V of AC power, and the battery supplies 12 V of DC power. It's not really a matter of simply "oh, 110 is bigger than 12, guess I'm screwed". You could use a power inverter, though I don't think power inverters are FIRST legal...

Have you written any successful programming logic since your last post?

I honestly have not done any programming since summer.

[PAR_WIG1350]

Originally Posted by davidthefat

I honestly have not done any programming since summer.

When an artist works on something for too long, nothing about it seems to be good enough. It isn't until he leaves it alone for a while before returning that he can see what truly needs to be worked on.

--Me

[theprgramerdude]

Agreed with the above. I backed off of it, and instead have been studying the bigger picture of how to build a decent robot, as well as some of the sensors available for input. There's no reason to try all autonomous anyways if your robot just flat-out sucks.

[davidthefat]

Well I plan to get a hold of that GadgetPC sometime soon and start working on the cameras and stuff. I really wanted to try parallel processing with multiple cores, but I guess I can hold off for a while.

[PAR_WIG1350]

Originally Posted by davidthefat

Well I plan to get a hold of that GadgetPC sometime soon and start working on the cameras and stuff. I really wanted to try parallel processing with multiple cores, but I guess I can hold off for a while.

Stick with what you know, especially that which you know but don't know that you know. Think about how you think. How do you tell what an object is? How do you know if it needs to be avoided? In humans, image processing occurs in the occipital lobes of the brain which communicate with the frontal lobes to determine what everything is based on stored information. In your system, the GadgetPC seems to be equivalent to the occipital lobes of the brain. The frontal lobes would be the CRIO, possibly the FPGA. This involves a lot of boolean logic.

is it a wall?
yes
{avoid}
no
{is it a robot?
   yes
   {do something, maybe add an extra "is it an opposing robot?" test}
   no
   {is it a scoring object?
        yes
        {pick it up}
        no
        {is it a goal?
             yes
             {score,  if scoring objects are in possession}
             no
             {is it part of the field that can be driven over?
                   yes
                   {ignore}
                   no
                   {avoid}
}}}

Another important part of the brain is the parietal lobe--> processes other sensory information and builds maps of the environment

Motion would be controlled by another part of the system that uses all of this information gathered by the three "lobes" and maps out the best route to take.
-----------------------------------------------------------------------------------
**note** this isn't exactly how the human brain works, I'm just simplifying it to fit the application better and to avoid confusion of some people, myself included.
___________________________________________
I'm sorry if this is hard to follow or makes no sense, I had a hard time wording it, or even figuring out what I was trying to say, maybe I should take a break and come back to it later

[davidthefat]

Originally Posted by PAR_WIG1350

Stick with what you know, especially that which you know but don't know that you know. Think about how you think. How do you tell what an object is? How do you know if it needs to be avoided? In humans, image processing occurs in the occipital lobes of the brain which communicate with the frontal lobes to determine what everything is based on stored information. In your system, the GadgetPC seems to be equivalent to the occipital lobes of the brain. The frontal lobes would be the CRIO, possibly the FPGA. This involves a lot of boolean logic.

is it a wall?
yes
{avoid}
no
{is it a robot?
   yes
   {do something, maybe add an extra "is it an opposing robot?" test}
   no
   {is it a scoring object?
        yes
        {pick it up}
        no
        {is it a goal?
             yes
             {score,  if scoring objects are in possession}
             no
             {is it part of the field that can be driven over?
                   yes
                   {ignore}
                   no
                   {avoid}
}}}

Another important part of the brain is the parietal lobe--> processes other sensory information and builds maps of the environment

Motion would be controlled by another part of the system that uses all of this information gathered by the three "lobes" and maps out the best route to take.
-----------------------------------------------------------------------------------
**note** this isn't exactly how the human brain works, I'm just simplifying it to fit the application better and to avoid confusion of some people, myself included.
___________________________________________
I'm sorry if this is hard to follow or makes no sense, I had a hard time wording it, or even figuring out what I was trying to say, maybe I should take a break and come back to it later

I am assuming the bumper system is going to be mandatory again, so I thought that I can just do a color detection of the blob and decide with that info if that robot is an enemy or not. Also for detecting the objects, I have to think a bit more about that. I thought that just comparing colors is simple enough, but that can be very shoddy.

I been looking at the Machine Learning lectures by Andrew Ng @Stanford, I think that will give me a better insight on this.

I totally understand what you are saying, my mind thinks the same. What I was hoping I could do was use the PS3, but that does not seem likely.

There are 8 SPEs, only 7 are available, but thats fine. I was thinking each SPE was to be responsible for one part. All the SPEs can access the images without writing to it, so no problem there since all the SPEs would be only reading. Like 1 SPE can do the color detection, another do the distances of the objects, another do object recognition, and ect. They all relay that info to the PPE which will then compile the info and then do the logic. Then the PS3, through the ethernet, sends the instructions to the cRio.

Now that seems like a stretch, but honestly I like to aim high. I am 1/4 through the MIT PS3 lectures, I have learned so much just from that LOL.

If the PS3 is legal and the DC to AC inverters are legal, I can go ahead with this. Only problem, its my only PS3, do I want to potentially risk it getting crushed or something? I have 2 options too, run linux or go the homebrew way? I got a 60GB Japanese launch PS3, I still run linux on it, I have not updated it. The linux libraries for the PS3 are well documented and very thorough (IBM wrote them), but I would assume the homebrew route would have incomplete and shady libraries since its in its infancy. The down side of the linux on PS3 is the boot time, takes at least 45 seconds to boot up. The game OS only takes 2 seconds.

[apalrd]

Originally Posted by davidthefat

... takes at least 45 seconds to boot up.

That's about as fast as the cRio anyway (seriously, it takes forever to boot)

[Andrew Schreiber]

Originally Posted by apalrd

That's about as fast as the cRio anyway (seriously, it takes forever to boot)

Which confuses me since the CRIO I have at work boots up and runs my code nearly instantly.

[davidthefat]

Originally Posted by Andrew Schreiber

Which confuses me since the CRIO I have at work boots up and runs my code nearly instantly.

I think he meant the time it takes to connect to the classmate and have full control of it. Usually our robot has a 10-15 second delay when the green light comes up on screen and when it actually has connection

[Andrew Schreiber]

Originally Posted by davidthefat

I think he meant the time it takes to connect to the classmate and have full control of it. Usually our robot has a 10-15 second delay when the green light comes up on screen and when it actually has connection

Which is not at all what he said. In the future we should try to be more precise on what we mean especially when complaining about things.

[biojae]

Originally Posted by PAR_WIG1350

Another important part of the brain is the parietal lobe--> processes other sensory information and builds maps of the environment

That's where SLAM (Simultaneous Localization And Mapping) can come in.
(or, a partial implementation seeing as the field doesn't change from one match to the next (hopefully),
and a full map could be made once)

Originally Posted by PAR_WIG1350

Motion would be controlled by another part of the system that uses all of this information gathered by the three "lobes" and maps out the best route to take.

With a full map made, a path planning algorithm (such as A*, Dijkstra's algorithm, or others) could find the best path (shortest, least time, less obstacles, etc.).
And if this was constantly updated with current sensor information, it could try to find a route around other robots too.

Originally Posted by davidthefat

I am thinking of using 2 60fps cameras as a stereo vision system.

Then the mapping phase would be fairly easy, as stereo vision can give you depth information.

This video shows an IRobot Create with a Kinect camera performing SLAM:
http://www.youtube.com/watch?v=dRPEns8MS2o

[ptan]

For what it's worth, you may want to check out CMU's intro to robotics course

http://www.cs.cmu.edu/afs/cs.cmu.edu...1/www/current/

Especially check out Lab3 which has the some useful dead reckoning code (Yes, we are using their code in our robots).

[davidthefat]

Originally Posted by ptan

For what it's worth, you may want to check out CMU's intro to robotics course

http://www.cs.cmu.edu/afs/cs.cmu.edu...1/www/current/

Especially check out Lab3 which has the some useful dead reckoning code (Yes, we are using their code in our robots).

I purposely did not want to use the CMUCams because I honestly want to experience the process of making such algorithms. Because this is not all about winning. I first took this challenge to push myself and really dedicate and learn. I am also purposely staying away from OpenCV and other libraries for the same reason. Yes I have read in a lot of posts about why not to reinvent the wheel. I do it because I love doing it, it builds character and you learn a lot more than just reusing what some one has made. I push myself now so I don't have to push my self later, I wish I had learned earlier on. I wish I have learned it in elementary or middle school because that really is a valuable thing to learn. When I am doing my graduate work, the things I work on will not be canned projects like the ones we do in science class, I would be the forefront of my area of expertise. If I learn to do it on my own, its alot better than learning it too late. Its different from just plugging in the variables in a equation and getting the right answer than understanding why and how you got that answer. I believe the latter is what I strive for.

People will disagree with my mentality especially engineers, but I feel that is what I need as a student and as a person. My counselor told me that no one will just wait for me to help me, I have to actively seek help if I need it. I always think that if you reinvent the wheel, chances of you doing it better is greater than just reusing it. If you just reuse it, you will never be able to improve it.

Also if I fail, I learn, which in my book is success.
"I have not failed. I've just found 10,000 ways that won't work." -Thomas Edison