Autonomous Robotic Platform Training on Behavioral Cloning Neural Networks using ROS and VESC Project Resources

Abstract

The endeavor on this paper aims to present a physical robotic platform with the main purpose of addressing the subject of autonomous mobility for both people and goods. The platform, which runs on ROS (Robot Operating System) and VESC Project resources, contains a basic implementation of a behavioral cloning deep neural network in order to achieve a certain level of autonomy, which will be further developed in a series of subsequent papers. Use cases for the autonomous platform are warehouse commodity management, medical material transportation in hospitals, airport luggage logistics and personal mobility for handicapped people. The progress landed in this regard goes in hand with self-driving cars, another key target use case for the proposed platform as test bench. Mobility tests are carried out to assert adequate physical operation, resulting in effective performance at up to 17km/h and secure current, voltage and temperature values for the brushless DC motors, battery and controllers. As for artificial intelligence testing, training accuracy for the neural network presents a value of 0.9536, whereas validation settles at 0.9481, which provides a confident trained model for later implementation.