A Sample Research Project Idea for Brain Controlled Cars

Training and Model Creation

- What does the model has to adapt to?

The model has to adapt to the brain state of "slight left-ness" and "slight-right-ness" of the brain, while overcoming the obstacle. This is not the regular large right turn or large left turn. But the vehicle will accumulate multiple small left turns/right turns to overcome the obstacle. This "slight left turn/slight right turn phenomenon" should be adapted by the model.  Two different models will be created: One to overcome the object from the left and another model to overcome the object from the right. 

Simulator Setup

- Multiple static objects of different shapes and sizes will be placed on the road. Place "Distance Markers" at the beginning and end of the "direction changes", while maneuvering the obstacle. Place a few "Distance Markers" over the straight drive. This will define the "beginning" and "end" of data collection snippets for left/right/straight drives. Drive the car such that the driver will overcome the objects from the left/right (A separate run for left drives and a separate run for right drives). Collect the data and create two models (one for left and one for right) by extracting the snippets. 

Classification

Collect the data for "left" run of the simulator

- Extract the snippets (These snippets will be of variable size, because the obstacles are of variable size. The only common thing is the "left-ness" in the turns). If possible use "sliding window" mechanism to classify the 'left" windows. And gather some data snippets from the "straight" drive. Assign a label "1" to "left" and label "0" to "straight" turns.

-  Repeat the above process for "right" run of the simulator to extract "right" and "straight" snippets. Assign label "2" to "right" and label "0" to "straight" turns. 

- Create a ML model with CNN. 

Real Time use of the model on the simulator

- One static object of any shape and size will be placed on the road. The driver will be instructed to maneuver the object either from "left" or from "right". On the occurrence of the obstacle, the car should automatically start turning slight left for "left side model' or slight right for "right side model". This is because, the brain signals are modelled that way. Both the experiments (of two models) will be conducted separately.