shivangg / rover_challenge

Based on the NASA Sample Search and Return challenge

Project: Search and Sample Return

Working GIF

Checkout the test video of this Rover challenge on my YouTube playlist.

Notebook Analysis

1. Ran the functions provided in the well designed Jupyter Notebook that helped even afterwards to rapidly design solutions for the autonomous driving. Defined functions for:

1. Navigation thresholding
2. Obstacle thresholding
3. Rock Sample thresholding

2. The process_image() function is populated with the appropriate analysis steps to identify the pixels corresponding to

• Navigable terrain using color_thresh_navigable

• Obstacle (inversion of the navigable terrain) using the obstacle_thresh_obstacle

• Golden Rock sample using the color_thresh_golden to find the pixels within the RGB range of the golden rocks.

Finally, map them onto the ground truth map of the world. The output images from the process_image() are used to create video output using the moviepy functions.

Autonomous Navigation and Mapping

1. The perception_step is derived from the process_image function in the Notebook. It gave a 30% improvement when I considered only a cropped out middle portion of the vision_image rather than the whole image.

    
h, w = navigable.shape[:2]
delta = 50
region_of_interest = navigable[h - delta: h,\
 						int(w/2) - delta : int(w/2) + delta ]

navigable_x , navigable_y = rover_coords(region_of_interest)

The decisions made by the rover for autonomous driving:

1. Right wall following by setting an offset to the steering that depends on the mean of nav_angles

steer_offset = -15
Rover.steer = np.clip(np.mean(Rover.nav_angles * 180/np.pi + steer_offset) \
					, -15, 15)

2.Detect if the rover is stuck by checking if the Rover state variable Rover.stuck if above the defined threshold Rover.stuck_time_threshold and start turning right to get out of stuck state.

if Rover.vel < 0.1 and Rover.vel > -1.0:
    Rover.stuck_time_threshold += 1
    Rover.throttle = Rover.throttle_set
    if Rover.stuck_time_threshold > 100:
        Rover.stuck = True
    if Rover.stuck:
        Rover.throttle = 0
        # Set brake to stored brake value
        Rover.brake = Rover.brake_set
        Rover.steer = 0
        Rover.mode = 'stop'

3. Fine tuned the constants for for better turning and faster controlling of the Rover

self.stop_forward = 150 # Initiate stopping sooner to prevent bumping into obstacle
self.go_forward = 400 # Threshold to go forward again
self.throttle_set = 0.6 # Increased throttle setting when accelerating for faster navigation

4. Added code in the supporting_functions.py for debugging purposes.

cv2.putText(map_add,"  Nav_angle len: "+str((len(Rover.nav_angles))), (0, 155), \
            cv2.FONT_HERSHEY_COMPLEX, 0.4, (255, 255, 255), 1)
cv2.putText(map_add,"  Mode: "+str(Rover.mode), (0, 145), \
            cv2.FONT_HERSHEY_COMPLEX, 0.4, (255, 255, 255), 1)

2. Launching in autonomous mode the rover can navigate and map autonomously. It can be improved by using:

1. HSV thresholding.

2. Using A* for path planning.

3. Better way to manage the state of the Rover.

4. PID tuning for the constants threshold_set, go_forward, stop_forward.

Specifications to reproduce the results:

• Resolution: 1024 x 768

• Graphis quality: Good

• Mapping: 99.9%

• Fidelity: 75.1%

• Time Taken: 420 seconds @ 20 Frames per second

The Unity simulator can be downloaded for Linux and Windows