Robotic movement prediction
• Dear all,

I've recently came across the Eureqa software.Seems to be an amazing product.
I teach robotics in a professional school.

One of the students project is to develop a centipede robot.
It has 13 "legs" and it "crawls" on a random based mode.

Recently a student asked me if it was possible to predict the centipede legs movement based
on the recorded data that we have (about 1200 "steps")
With this challenge in hands I decided to give a try with Eureqa.

The collected data is in the following arrangement:

Step. #1   1,0,1,1,-1,0,0,1,-1,1,-1,1,1
Step. #2   etc etc.

Each step has 13 legs status that can be:
forward (1),
stay in the same position (0)
move backward (-1)

My question to whom might help is: what will be the target expression to be inserted in Eureqa in order to find the next 13 legs "step"?

Thanks in advance.

Joao Bezouro
 Tweet
• Hi Joao,

This sounds like a great use case. Have you looked into Eureqa's time series tutorials yet? The way you have the problem set up, each column represents a leg and each row represents a step. You'll have to predict each leg individually, so you'd set up 13 different searches with 13 different target expressions something like this:

Search 1: leg1 = f( delay(leg1,1), delay(leg2,1), ..., delay(leg13, 1) )
Search 2: leg2 = f( delay(leg1,1), delay(leg2,1), ..., delay(leg13, 1) )
...
Search 13: leg13 = f( delay(leg1,1), delay(leg2,1), ..., delay(leg13, 1) )

Then be sure to include all the history building blocks (and perhaps logical as well) on the Define Search tab. Once you've found promising solutions for each leg, you can evaluate those solutions on new data using the "Evaluate model values from a dataset" tool in the Reports tab.

Thanks,
Jess
• Hi Jess.

Thank you for the input
This approach seems to be the best one (I'm a beginner so any valuable opinion is certainly welcome).
Me and the students will give it a try.

If we experience difficulties we'll drop a line here.
Thanks again for your help

JBezouro

#### Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!