Domain Adaptive Imitation Learning for Self Driving Environments
To overcome this challenge, we propose a recursive adaptive prediction but due to time limitations, we will defer its implementation to future work. This approach necessitates the model to predict the next position recursively based on the forecasted location in the previous time step, optimizing the model over the entire trajectory. By doing so, the model will be forced to consider all the input information, leading to improved performance and a better understanding of the driving policy.
E.H.Z. Al Suradi, "Domain Adaptive Imitation Learning for Self Driving Environments", M.S. Thesis, Machine Learning, Abu Dhabi, UAE, 2023.