Path planning for a robotic vehicle using imitation learning

In this project I used color information from an aerial map of a city to plan the path between any two points for a robotic car and for a robotic pedestrian. A cost map was learned based on color and edge information using imitation learning i.e. 20 paths for car and pedestrian were specified by hand and the algorithm built a cost map that was as consistent as possible with all these paths.

![car_cost] (car_cost.jpg)

(Above) The cost map learnt for cars (blue = low cost, red = high cost)

![ped_cost] (ped_cost.jpg)

(Above) The cost map learnt for pedestrians (blue = low cost, red = high cost)

Using these cost maps, paths were planned between any two arbitrary points on the map for car / pedestrian:

![car_path1] (car_path1.jpg)

![car_path2] (car_path2.jpg)

![car_path3] (car_path3.jpg)

(Above) Paths for cars

![ped_path1] (ped_path1.jpg)

(Above) Path for pedestrian

See the project report in the 'info' folder for more details.