Training the tcp model without policy model.

[adnanfidan]

I am currently working on integrating between the Donkey Simulator and the TCP model. I mean to apply the TCP system using another simulator(Donkey), for a lane keeping task. I’ve been using this setup for my experiments. When I was collecting the data, I noticed that the dataset normally includes mu branches and sigma branches. As I understand it, these are two parameters derived from the policy model. I’m using the donkey version and working with a simpler scenario, so I preferred to collect data with a pure pursuit algorithm. But this method doesn’t let me work out the mu and sigma values.

Is it possible to calculate mu and sigma without using the policy model or would it be feasible to adjust the model so that it can still
produce actions when these parameters are omitted? Thank you in advance.