my question is related to Neural Network architectures and how to handle a specific situation. Please bear with me, I'm studdying this by myself, reading books, taking online courses and just got to this situation:

I have a simple project where a vehicle moves around on a 2D plane. This vehicle must avoid collisions with other objects while it moves from one place to another.

I first designed an "avoidance" behaviour with vectors/forces -based on Reynolds work. This routine I made will return the direction and length of the vector the vehicle should take to avoid a single object. Then I generated data from different random scenarios and trained a neural network with it.

This seems to workd great!

This are my INPUTS and OUTPUTS:

[vehicle_x, vehicle_y, vehicle_velocity_x, vehicle_velocity_y, obstacle_x, obstacle_y, obstacle_velocity_x, obstacle_velocity_y]

[ x, y, length ]

The problem here is that I can only work with one obstacle at a time and if I wanted to take a different decision for special scenarios I can't. For example, if many obstacles are near each other by a certain distance I would like to treat them as a "group" and calculate angle and force in a slighlt different manner.

So, the question is, what architecture or approach can I use to be able to feed a network with one or more obstacles?

Thanks in advance!

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.