I am working on training the TCP (Trajectory-guided Control Prediction for End-to-end Autonomous Driving: A Simple yet Strong Baseline) model for lane keeping tasks. In the original setup, the data collection process relies on the carla roach model, and the returned mu and sigma branches values are then used to calculate the actions. In my case, since I am focusing only on lane keeping and not using CARLA with the carla roach model, I prefer to compute actions mathematically with a pure pursuit algorithm during data collection. However, this approach does not give me access to the mu and sigma values.
In short, I would like to better understand
1- the logic behind the mu branches and sigma branches parameters that are returned by this model.
2- is there a way to calculate mu and sigma branches mathematically** without using the Roach model?
thank you in advance
I am working on training the TCP (Trajectory-guided Control Prediction for End-to-end Autonomous Driving: A Simple yet Strong Baseline) model for lane keeping tasks. In the original setup, the data collection process relies on the carla roach model, and the returned mu and sigma branches values are then used to calculate the actions. In my case, since I am focusing only on lane keeping and not using CARLA with the carla roach model, I prefer to compute actions mathematically with a pure pursuit algorithm during data collection. However, this approach does not give me access to the mu and sigma values.
In short, I would like to better understand
1- the logic behind the mu branches and sigma branches parameters that are returned by this model.
2- is there a way to calculate mu and sigma branches mathematically** without using the Roach model?
thank you in advance