![]() ![]() Inducing a first-level training of the model in a simulated world allows the agent to explore and generate collision leading scenarios. We take a baseline imitation learning model without transfer learning and our model that exhibits transfer learning, then compare their performance to demonstrate that transfer learned models are able to handle new scenarios much better. We prove our hypothesis by utilizing imitation learning architecture that comprises of sequential neural network design, similar to that of the advantage-calculating neural networks of the PPO method. We propose that both of these challenges of the policy gradient approach can be addressed by inducing transfer learning to the individual components of the policy gradient approach.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |