paper: Adaptable Deadband with Perceptrons

[E Dawg]

Thread created automatically to discuss a document in CD-Media.

Adaptable Deadband with Perceptrons by E Dawg

[E Dawg]

What is it?

A deadband that changes itself based on the user input using perceptrons. For example, if the driver is constantly providing maximum input, the deadband change to increase the threshold. If the driver is providing low input, the deadband will change to decrease the threshold.

What are perceptrons?

Perceptrons are a form of learning AI that mimics a human brain cell. It can take any number of inputs. It then modifies each input by an individual weight. This weight is what is modified as the perceptron learns. The weighted sum is then passed through an activation function to either 'fire' or not. The equation for changing the weight is as follows:

W(n) = W(n+1) + η(d(n)-y(n))*x(n)
Above W(n) is the old weights vector, W(n+1) is the new weights vector η is a user-defined constant called the teaching step, d(n) is the target vector, y(n) is the actual output of the network and x(n) is the corresponding input.

Like a neuron, perceptrons can also be linked together. This deadband only uses one perceptron, however.

You can download the source code above.

[Jared Russell]

What problem is this a solution for?

[E Dawg]

Originally Posted by Jared Russell

What problem is this a solution for?

1. It saves time having to change code every time a driver says "the deadband is too sensitive / not sensitive enough".

2. Sometimes it is good to have a highly sensitive deadband (maybe you need to go really slow), and this way you are able to do that without having a super-sensitive deadband all the time.

3. It adapts to individual situations.

4. I thought it would be a fun summer project.

[Chris is me]

If I understand this correctly, this would change the response of the joysticks in real time (i.e. throughout a match)? I feel like this, while very cool, wouldn't be optimal simply because the same input at different times would produce different results. Practicing maneuvers and getting them to the point where you just use muscle memory to drive the robot would be pretty much impossible.

[AllenGregoryIV]

I haven't looked through the code but this sounds very cool. My only questions is what problem does this solve? Were you having control issues or something and this fixed them?