orientation sensors

today i spent a lot of time trying to find some sort of orientation sensor. you can buy really damn expensive vr trackers. the low end there is $1000 and it can go up a lot more because people that make vr stuff suck. you can also buy cool little solid state sensors and try to make them into something interesting. as usual that means a visit to the super awesome http://sparkfun.com who did most of the work for you. a 6dof setup can be had all ready to go for $350ish.

there seem to be a few basic pieces of tech in an orientation sensor.

gyro. this will measure a rate of change about an axis. i've only seen single axis sensors. they are quick, but they drift like a bitch because they are relative. everyone knows the devil invented relative systems. you'll need 3 of these aligned perpendicular. i've read that rc units are prettly slow because of all the pwm talking in and out. maybe 15ms reads. if you can grab the data direct from the ic you can get in the 500's of samples per second.

accelerometers. these measure an acceleration with respect to gravity. because there is a frame of reference they are absolute, but i really don't know how that is possible. i think they might be.. relatively absolute? anyway. you need 3 of these perpendicular to each other. you can however buy a part that has all 3. so these can be used to measure tilt, but they are slower than gyros. unfortunately.. since they are acceleratometers.. they can be fucked with via fast translations too! i'm not sure if that is an issue or not.

there also are magnometers that measure.. guess? those however i've read have trouble with motors interfering with them. servo = motor. so no go.

so you take a 3axis accelerometer and 3 gyros. use the gyro for your fast updates and use the accelerometer to fix your errors.

right now i'm still undecided.

maybe i should look into positional tracking and convert it to an orientation. unfortunately that requires some sort of reference base station. i could use a vision solution. i ran across this.. www.naturalpoint.com ..not the game stuff, the optitrack sdk stuff. using 2 of those i should be able to track the robot pretty accurately, but of course you need the camera aimed at it. anyway.. for a vision camera that is pretty damn cheap. $250 for 120hz (it spits out 2d positions + pixel area, not the image). the nice thing about this is i could also use those cameras to build a motion capture studio and a face capture rig.

or.. maybe a better solution is to put a camera in the robots crotch aiming down with a tight fov. his legs should not get in the way too much. i could make a laser range finder (which is on my project list anyway).

maybe i can get away with a crap orientation sensor if i have semi-decent feet pressure sensors. ie.. like a human.

maybe what i really should do is not spend so much time in the physical world and get cracking on robotengineer instead.
[/url]