Mobile Robot System

So to do that you will need to create a custom kinematic that contains a mathematical model of how the linear cartesian axis coordinates in the gcode file are transformed into rotational positions of your four wheels.

linuxcnc.org/docs/html/motion/kinematics.html

Creating that file has been greatly simplified recently but you will need to come up with the mathematical model of the forward and inverse kinematic for your robot. I guess you would want to use parameters for things like wheel diameters and distance from one wheel to the other. Also, because your feedback is coming from a separate encoder wheel that is located in a different location you will have to factor that in as well.

Hi,

For this device, does it has a radiocontrolled position feeback for x,y, and rotation?
When out of position during cutting, how is this checked?
 

Radio controlled position feedback? (Sorry did not get this)
The communication between the bot and the control system is not wireless but hard communication lines and the feedback for locomotion is mapped through the two rotary encoders mounted externally. Is that what you were asking about?

What accuracy are you aiming for? Dead reckoning over 15 meters will likely produce a lot of error in the position feedback, even disregarding EMI problems and slippage. You're likely to have to resort to an external navigation system, like beacons of some kind and detectors on your robot, to get absolute position.

Aim is 1mm
I had thought about having external positioning loop but am focusing on first getting the system functional at 1mm accuracy.
One thought is to have intermediate homing/calibration. Say at 3m intervals. But I don't know how to implement this in LinuxCNC.

You absolutely will not be able to hold even close to 1 mm accuracy over 15 meters using dead reckoning. A single degree of directional error will lead to an inaccuracy of about 17 mm after moving just 1 meter.

You can use an IMU plus encoders to help your robot figure out its approximate heading, but you need continuous external feedback of absolute position to have a chance of getting millimeter accuracy. (And you will probably need to write code to use that feedback to update the robot's idea of its heading based on the results of previous moves.)

What solution would you rather propose? (Sensor wise. Like what external feedback mechanism would you suggest)
Do you know of ways to do this in ROS perhaps?

Is there a reason to not just put the encoders on the driven wheels? I'd be skeptical of relying on encoder wheels for positioning. There's a high risk of it slowly drifting off course due to dust/slipping of the encoder wheels. meaning when it travels down and back, the cuts will almost definitely not line up at the end. Some absolute positioning system would be best.

Such as a vision based system that tracks markers, but that's (likely) not worth the cost. it would however result in the least amount of physical stuff you need to move.

You could try getting cable encoders, put one on each corner, then attach all the cables to the robot. Then with some math you can calculate the robots position.

Alternatively, you can make a gantry style machine. A 3m long beam rides on 15m rails. It isn't nearly as cool as the robot that drives around but it's a lot simpler.

If you have the time and resources, I would look into the vision systems. Shaper makes a handheld cnc router that uses cameras for tracking. There might be a way you could do something similar.

If you have 2 of above devices. And a digital compass. I think it will work.
You have to unpack the devices and use the lenght input as input for x and y. Then the compass will tell the robot if the rotation value is correct.
This will be the most cost efficient solution.




 

dhruv wrote:

What solution would you rather propose? (Sensor wise. Like what external feedback mechanism would you suggest)

I think your best bet is to make your robot move in only one axis, and make it 3 meters wide - i.e. build a gantry on wheels!

I'm not aware of any local positioning system that gives millimeter accuracy. Ultrasound time-of-flight seems to be able to give 1 cm accuracy at most. Others, like UWB (ultra-wide band) are worse. I think you will be hard pressed to get 1 mm accuracy from a vision-based system, and the blazing plasma arc sure won't help!

LIDAR scanners also have a distance accuracy of about 10 mm (after statistical treatment of multiple measurements).