The project tackles the problem of multi-target domains/contexts for the self-driving car in urban cities, using the unsupervised domain adaptation (UDA) methodology. and compares two approaches:
1- `Baseline Model`: which deals with the problem of the multi targets domain as if it is a single target.
2- `MTKT Model`, which consists of multiple teachers [one for each target domain] that distill their knowledge to a single student which will be used as the main model in the inference time.
Our project built on the latest SOTA paper of #Valeo France https://arxiv.org/abs/2108.06962
Also, we have developed a self-driving car agent implementing the Deep #Reinforcement algorithm #DQN inside #Carla Simulator.