It is crunch time now. I am down to less than a week to go before the contest. Sorry I have not been good about posting lately. Rockhopper is completely built now (or as built as it is going to be). Building the bot itself and getting the low-level programming done and working took longer than expected (i.e. see my last post about the i2c performance issue). So, it looks like once again I don't have much time left to do the interesting high-level AI programming I wanted to do. I do however have it working fairly well now. It can detect and put out the candle without much trouble. It can also do a fairly good job of not getting stuck in the maze. I was having some trouble where it would run into the corner of a wall if it was in the blind spot of my ir range sensors and sometimes the bumper switches would not trigger so it would just keep pushing against the wall (very bad). I added some code to check the wheel speed based on the wheel encoders so now when the rpm of the wheels drops it knows it is stuck. This seems to be working nicely now.
I also built a beeper used to mark where the "baby" is in the maze. The "baby" is a 15 watt light bulb painted black. When the robot finds this it is supposed to drop a beeper that beeps 2 times per second. I am really happy with my solution for how I drop the beeper. I will try to take pictures of it if I get a chance. Basically it is just a motor on a hinge with a small rubber wheel on it. The wight of the motor holds the beeper in place then when the motor is turned on, the wheel spins the beeper off. This also releases the switch on the beeper which turns it on. This all works better than expected. I now need to work on my code for finding the baby itself. I plan on doing that later today.
What I have been working on the last couple of days is trying to use a neural network to interpret the sonar data. I hope that this will help filter out the false pings you commonly get with sonars. So far I am not having much luck but I have found a few critical errors in my code. I wrote the original code late at night long after I should have gone to bed so it isn't as clean as I would like. If I don't get it working somewhat well by lunchtime today I am going to scrap the neural net idea and write some simple code that uses the raw data. It will be more likely to do strange things because of false pings but at this point I am running out of time!
So, looks like I can certainly do the task of the competition but not as well as I would like. Probably the biggest thing that I need that I have done no work on is somewhat to know if it is just doing circles in the maze and not actually exploring new areas. If I get my sonars and the "baby" detection stuff working that will be my next task.
Next Tuesday I go back to work. I really am enjoying my time off but it will be nice to get back to a 5 day work week! I have been averaging about 10 hours a day 7 days a week on this.