Self Driven Car

Computer Science & Engineering

An adversarial network is a deep learning framework that makes use of multiple deep learning network as “adversaries” critiquing the results generated to maximize the probability of generative results. This can be employed in training a Self Driving Car. This cannot be embedded in a self driving car until fully trained in a simulated environment. The self driving car and it’s sensors are to be operated in a virtual environment. The model is so trained that an actor critic model M trains to manoeuvre the vehicle with a reward based system. A discriminator network uses it’s pre-trained models to critique the efficiency of the self driving car. The car is set in obstacle courses which it is to avoid. The unsupervised learning process if critiqued using a discriminator like a mentor to the network, until the self driving car learns to avoid all obstacles. Further the camera sensors of the car using YOLO(You only look once algorithm) to detect objects and noise based transformations are applied to understand the the range of objects to avoid and signs to read. Convolution neural networks are used for the YOLO algorithm to actively distinguish objects to avoid, road unevenness to reduce speeds, speed breakers to tackle and unpredictable human behaviour to account for.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.