Add Configuration Parameters for Robustness Experiments

[SakshayMahna]

In reference to issue #313

For Camera Position, a temporary model f1_temp is created in the Gazebo models directory. The tmp_world created by tmp_world_generator.py uses this f1_temp model. The example config file is default-multiple-robustness.yml.

For Noise, a new utility script added in utils folder. The noise is added to the camera image just before it is fed to the Neural Network. Example script is brain_f1_explicit_noise.py and example configuration file is default-noise.yml.

[sergiopaniego]

Nice one!
I have a question regarding the noise addition. Would it be possible adding it as another image transformation option? Following the current pipeline:

BehaviorMetrics/behavior_metrics/brains/f1/brain_f1_opencv.py

Line 161 in b446b4c

image = self.handler.transform_image(image, self.config['ImageTranform'])

What do you think about it?

Is it possible modifying the camera on execution with the graphical interface? You have only provided the multiple config file so I guess that it's only intended for scripting purposes.

[SakshayMahna]

Yes, we can add the noise to the current pipeline. The code would be much more organized in that way. In the albumentations library, there is one transformation for Gaussian Noise, but no such function for Salt and Pepper Noise. I can use my custom implementation for the same.

I'm not sure how to modify the camera on execution. But, we can see the noisy image in the graphical execution. How should I go about it? We can possibly do the same for all the transformations present in the handler.