DESIGN CHALLENGES
-
Capture useful data and removing unwanted data while training the car.
-
Integrating data from Radar, Lidar and Camera.
-
Adjusting epochs, batch size, learning rate and dropout to get maximum accuracy while training the car.
-
Preparing good model like adding Maxpool, RELU, Dropout to LeNet model architecture for good training accuracy.
-
Gather data in multiple angles, in reverse order, in corner cases to keep the car in the center of the lane.
-
Designing ROS architecture.
-
Calculating trajectory for smooth curve and jerk free change of lane.
-
Designing PID while coming to stop for traffic signals and another car or obstacle ahead.
MANTRA BEHIND SUCCESS
-
Skillful and persistent research work for 6 months.
-
Setting goals, meeting them regularly.
-
Willingness to learn new things when required.
-
Consistent use of trial and error method to improve the efficiency and accuracy.
-
Proper calculation helps depending upon the task to be done.
-
Start early and document as you go ahead, which helps later while improving code in later part while tuning its accuracy.
FUTURE CHALLENGES
-
Collect dataset in extreme weather conditions like Snowfall or Rainfall.
-
Self driving car should work smoothly in extreme weather conditions.
-
All weather localization and mapping.
-
Robust moving object detection.