I'm trying to learn the parameters of the car by myself. The inputs, outputs and neural network architecture are exactly the same as it is in the repository and in the paper. The data I have collected is from 60 minutes of various driving in a simulation.
I'm using PyTorch to train the network and get mean total errors between 0.2 to 0.3. When I plug the dynamics in the simulation again, the trajectory planned looks correct, however, the car steers heavily to one side (left) and fails to follow the trajectory. Then it realises it has missed it and just reverses to attempt it again and repeat..
Hello,
I'm trying to learn the parameters of the car by myself. The inputs, outputs and neural network architecture are exactly the same as it is in the repository and in the paper. The data I have collected is from 60 minutes of various driving in a simulation.
I'm using PyTorch to train the network and get mean total errors between 0.2 to 0.3. When I plug the dynamics in the simulation again, the trajectory planned looks correct, however, the car steers heavily to one side (left) and fails to follow the trajectory. Then it realises it has missed it and just reverses to attempt it again and repeat..