e2e_nav

new dataset, attack cases, vae, gat, rec, nav.

Functionality (completed)

1. Implement 6 types of attacks and visualize them for verification.

2. Implemented a 4 node attack.

3. Added equation 2 (step2). Fixed parameters of model_all.model_vae_fixed_parameters and implemented scalable model training, the fixed parameters were later added to model_net.py.

4. Simplifies the image semantic attack code.

5. Added red light input to the navigation model.

6. Add the test agent under carla.

7. Added 2 new attacks: repeat frame, fan masking.