Title

Applications of Machine Learning Algorithms In Lidar-Based Autonomous Vehicles

Author(s)

Daniel LiFollow

School Name

South Carolina Governor's School for Science & Mathematics

Grade Level

12th Grade

Presentation Topic

Computer Science

Presentation Type

Mentored

Abstract

Vehicle manufacturers such as Toyota and Ford plan to develop mostly-autonomous vehicles within the next decade. To ensure the utmost safety for all that share the road, the autonomous driving algorithms must be accurate. The goal of this research was to develop an accurate machine learning algorithm able to navigate paths using infrared laser data from Light Detection and Ranging, LiDAR for short, and to test its capability to learn from various training data. Automation was implemented in a remote-controlled car which consisted of various sensors including LiDAR, camera, and inertial measurement unit. Cardboard boxes were used to construct tracks for testing the control-based driving algorithm and gathering data for training the neural network. Trained neural networks suggest that diverse data consisting of both human driving and control-based driving provides the best training for machine learning.

Location

Founders Hall 140 A

Start Date

3-30-2019 9:00 AM

Presentation Format

Oral Only

Group Project

No

COinS
 
Mar 30th, 9:00 AM

Applications of Machine Learning Algorithms In Lidar-Based Autonomous Vehicles

Founders Hall 140 A

Vehicle manufacturers such as Toyota and Ford plan to develop mostly-autonomous vehicles within the next decade. To ensure the utmost safety for all that share the road, the autonomous driving algorithms must be accurate. The goal of this research was to develop an accurate machine learning algorithm able to navigate paths using infrared laser data from Light Detection and Ranging, LiDAR for short, and to test its capability to learn from various training data. Automation was implemented in a remote-controlled car which consisted of various sensors including LiDAR, camera, and inertial measurement unit. Cardboard boxes were used to construct tracks for testing the control-based driving algorithm and gathering data for training the neural network. Trained neural networks suggest that diverse data consisting of both human driving and control-based driving provides the best training for machine learning.