SLAM without motor encoders? *

[vidam]

We are working on a Java class called Motor_Util.java and the methods utilities we will need. The self-modified Pioneer does not have any encoders unfortunately because of the setup we have using the Sabertooth motor driver and the RoboRealm software. The motors are DC brushless with built-in encoders. But we have no idea where to connect the encoders given the current hardware setup. Is SLAM even possible without encoders and just using a webcam, GPS, and sonar to do localization and mapping? We are given waypoints right before the competition so the only landmarks we can localize on are roads and maybe some pre-stored images or map of the trail.

[david_or_johnny]

I am not sure, but I was wondering if you were going to write a Rao-Blackwell particle filter or an extended Kalman filter. Thanks. I am trying to slap together my own SLAM with only 3 sonar sensors, a compass, and an encoder.

[paulstreats]

im trying to put one together without any position of reference. where the bot takes one scan, then takes another. If the 2 scans overlap to a certain extent then they will become fused together in the master map. The robot can find its position by referencing close objects

[sdk32285]

I don't know your time frame but you can do visual odometry with the web cam...

Do you have an IMU (or are you on a flat floor where you can make assumptions that the roll and pitch are constant, and that heading is found by looking at past GPS points), if so you can just align each sonar scan to overlap the previous scan based from the distance traveled and your heading.
This way you can close the loop by using GPS and if your ambitious you can do some scan matching on your map to confirm that you actually closed the loop.

[vidam]

I don't know your time frame but you can do visual odometry with the web cam...

Do you have an IMU (or are you on a flat floor where you can make assumptions that the roll and pitch are constant, and that heading is found by looking at past GPS points), if so you can just align each sonar scan to overlap the previous scan based from the distance traveled and your heading.
This way you can close the loop by using GPS and if your ambitious you can do some scan matching on your map to confirm that you actually closed the loop.

My time-frame is early Spring, April. The landscape is comprised of rocks, gravel, grass, and road. Did not even consider IMU (mainly because I'm new to this kind of sensor technology for robots). Thanks for the ideas!!!! I have a 45-90 degree ramp of work ahead of me.

[sdk32285]

Hi
So an IMU will give you your RPY (roll,pitch,yaw) that you will need to acuratly patch togethor the scans. However they can be pricey. I f you can ignore the effects of the rocks you might just want to use a digital compass for your heading.

[vidam]

Hi
So an IMU will give you your RPY (roll,pitch,yaw) that you will need to acuratly patch togethor the scans. However they can be pricey. I f you can ignore the effects of the rocks you might just want to use a digital compass for your heading.

You hinted at the $$$ for IMU? Couldn't one just buy a couple of cheap accelerometers and a gyro and mock up the IMU? How much are they exactly $$$-$$$$?

[Admin]

You hinted at the $$$ for IMU? Couldn't one just buy a couple of cheap accelerometers and a gyro and mock up the IMU? How much are they exactly $$$-$$$$?

time is money . . . and depends on the error you don't mind getting . . . google to see IMU pricing

[vidam]

Hi
So an IMU will give you your RPY (roll,pitch,yaw) that you will need to acuratly patch togethor the scans. However they can be pricey. I f you can ignore the effects of the rocks you might just want to use a digital compass for your heading.

I was just reviewing an old thread I started...and i ran across SDK32285's recommendation to use a digital compass. And it finally sunk into my brain that this is probably the better alternative. Thank you SDK32285!

[msk]

ratSLAM is a SLAM system using a Pioneer robot with a camera.

http://www.itee.uq.edu.au/~milford/RatSLAM/RatSLAM_main.htm

If your university has a site subscription you can access a book/phd thesis on ratSLAM here: http://www.springerlink.com/content/g5g1u353t743/

[ArcMan]

Just in case you haven't seen them yet, Sparkfun is selling some inexpensive IMU's - about $325.

[msk]

RatSLAM is actually based on a neural network which is inspired by the way that the biological navigational neural networks work in the hippocampus of rats.

Here is another webpage on RatSLAM, http://www.itee.uq.edu.au/~wyeth/RatSLAM/

There is some RatSLAM source code at Google Code: http://code.google.com/p/rsnxt08/source/browse/

MonoSLAM is another vision based SLAM system: http://www.doc.ic.ac.uk/~ajd/Scene/download.html

[Half Shell]

If you are trying to figure out how to record the speed of a motor without encoders, look into "back emf". By using a simple ADC to measure this, you can easily create a velocity feedback control and determine the resistance on the motor and its current speed. From this you can determine total distance traveled, albeit with more calculations than encoders.