I have a set of scenarios which represent the movement of a car in a certain environment containing some obstacles. So for each scenario I have the position of the car (x,y,t) and a description of the environment (a grid where holes to represent the presence of obstacles). Therefore for each element of the training set I have:

  1. label: List of car positions for each time instant (basically a trajectory)
  2. features: List of grids describing the environment around the car for each time instant (changing environment)

My goal is to use some machine learning tool to teach it how the car should behave given a new sequence of environments (list of grids).

What machine learning tool should i use? Can you provide some example similar to what I am trying to do in order to have some inspiration?

  • $\begingroup$ Is the goal of the car to reach a specific cell? $\endgroup$
    – zachdj
    Oct 1, 2019 at 13:18
  • $\begingroup$ @zachdj good question... yes the final point should be a specific cell of the grid $\endgroup$ Oct 1, 2019 at 18:43
  • $\begingroup$ In that case, it sounds like reinforcement learning is your best best. Are you able to run new simulations of the car, or are you restricted to the examples in your training set? $\endgroup$
    – zachdj
    Oct 1, 2019 at 19:03
  • $\begingroup$ @zachdj I thought about reinforcement learning but my algorithm should learn from the behaviors training set $\endgroup$ Oct 2, 2019 at 8:27
  • $\begingroup$ If you absolutely must learn from the training set, then it will restrict what type of algorithms you can run. Here's a pretty good article about learning from a "replay buffer": towardsdatascience.com/… $\endgroup$
    – zachdj
    Oct 2, 2019 at 12:59


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.