| Abstract [eng] |
The work deals with the design of an autonomous interface using Neural Network that helps a mobile robot to navigate in space autonomously. This paper presents a unique design for a autonomous interface using Behavioral Cloning for the process of training the robotic-system, to teach the robot how to navigate in space and avoid obstacles. Designing of AISRA was an end to end process, this document holds the details of data accumulation, preprocessing of data, neural network architecture and inferencing. Unlike the existing camera-mounted stimulated model, AISRA design is based on the laser scan information and leveraging the power of neural networks. The analysis study is about analyzing laser scan information for input to the system augmented with Neural Network instead of the existent camera-mounted stimulated models. In a traditional behavioral cloned system, training data set fed to the system will contain recorded decisions a human will make to turn the steering wheel to overcome the obstacles ahead. So, a study is conducted using a Robot Operating System based mobile robot Turtlebot 3 Burger, to run in a real-time environment accumulate to gather the data. Turtlebot 3 comes with a Laser Distance Sensor that when used for SLAM operations which helps us to gather Laser Distance Range data for 360-degree circumference. The system contains Keras deep learning library used for architecting the neural network and evaluating its performance. The final trained model is transferred to Nvidia’s Jetson to inference the data in real time. |
