Classification and Detection of Obstacles for Rover Navigation

Abstract

In this research project, the author aims to achieve Level 3 conditional automation whereby the researched Unmanned Ground Vehicle (UGV) is bound to classify and detect its own obstacle with human assistance as it cruises through a plantation field. Recognizing the different classes of obstacles enable the UGV to plan out the most efficient path to meet its desired goal. The purpose of this research project was to develop a classification and detection of obstacle and an optimal path planning algorithm suitable to be implemented for relieving the working process in an extreme condition plantation field. This paper presents an algorithm whereby it can conduct image-based obstacle detection through image masking and model prediction, along with a trigonometrical-based path planning approach. The proposed algorithm should hypothetically allow the UGV to conduct real-time path planning as it classifies some common obstacles such as leaves, rocks, and branches existing in a plantation field. As the waypoints were marked from the Ground Control Station (GCS), the UGV will travel towards the given waypoints to complete the given mission. When the UGV meets an obstacle, it will first differentiate whether it's traversable, followed by running the proposed algorithm to avoid the risk of destructing the UGV by choosing a collision free path. The basic idea is to apply path planning by considering the available spacing between the detected obstacle by comparing with a predefined threshold. Through the provided threshold value, the UGV can identify the type of obstacle yet to be detected. For instance, obstacles within the given range of value can be labelled to be a leafy obstacle, otherwise it is not considered to be a leafy obstacle. To ensure the behaviour and safety measure of the UGV to run smoothly, the author had undergone model training for an elevated model prediction by training and deploying a custom training loop through TensorFlow. Nevertheless, MATLAB was utilized to test out the concept of the path planning algorithm to examine its behaviour as untraversable obstacles were met. All these implementations can further grow in the agricultural industries as it can aid humans with performing tedious and impossible tasks on site.