detail explanation needed

[ 30 points] Alice and Bob work on a self-driving car project. They want to develop a generalized model that can drive in most roads safely. Bob has trained a Markov Decision Process based model (MDPReinforcement Learning) on a dataset with 100,000 samples. [Provide as much details as possible for your answers] a. [15] Alice looks for better evidence to convince Bob that his trained RL-MDP based model is not a robust self-driving technique. She decides to use the trained model on a partly marked and semi-chaotic road environment. What does she need to do convince Bob of her hypothesis? b. [15] After seeing the results of Alice's experiments, Bob decides to retrain the RL-MDP based model in a way that the model would be robust. Suggest two different solutions for improving the robustness of his RL-MDP based model