Self-Driving Car: Simulation of Highly Automated Vehicle Technology using Convolution Neural Networks

Engineering notes

The proposed model has been evaluated considering the-No of epochs vs loss calculation, as performance metrics, and was found that the proposed model has shown superiority with the existing works.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

Driver behaviour is a significant factor in the smooth driving of vehicles on the roads. 94 % of crashes and road accidents are prone to drivers' rash driving behaviour. To address issues related to road accidents and crashing of vehicles on the road, Highly Automated Vehicle (HAV) Technologies have been proposed. Self-Driving Cars are part of Highly Automated Tech-nologies having promising benefits ranging from Greater Road Safety, Greater Independence, Saving money, More Productivity, Reduced Congestion and Green House Gains. The current study focuses on the deployment of self-driving automobiles based on the Deep Learning paradigm. The automobile has been simulated on the Udacity simulator for convenience and safety. On the Udacity platform, a technique for training and simulating an unmanned vehicle model using a convolutional neural network has been developed. The data used to train the model is captured in the simulator and fed as input into the Deep CNN. Following data collection, Deep CNN is trained to have Safety Navigation by passing Steering, Throttle, Brake and Speed as Control Inputs. The use of three cameras considerably improves the precision of the navigation job. To manage the car, the steering wheel amount will be modified such that it runs in the centre of the lane. We evaluated the model using UDACITY's simulation system. The proposed model has been evaluated considering the-No of epochs vs loss calculation, as performance metrics, and was found that the proposed model has shown superiority with the existing works.

Links and sources

• Official / arXiv page

Comments