Machine Learning based Intelligent Model for Path Planning Obstacle Avoidance in Dense Environments for Autonomous Mobile Robot

Abstract

Abstract In this paper, a unique Machine Learning (ML) model namely, Adaptive Block Coordinate Descent Logistic Regression (ABCDLR), is proposed for segregating the movement of an Autonomous Mobile Robot (AMR) by framing it as three class problem, i.e., no, left, and right turn. The velocities of the left and right wheels, as well as the distance of the obstacle from AMR, are collected in real time by two Infrared (IR) and one Ultrasonic (US) sensors, respectively. The performance of the proposed algorithm is compared with three other state-of-the-art ML algorithms, such as, K-Nearest Neighbour (KNN), Naïve Baiyes, and Gradient Boosting, for obstacle avoidance by AMR; considering the accuracy, sensitivity, specificity, precision values for three different speed conditions, i.e., low, medium, and high. Various Logistic Regression (LR) model parameters, such as, pseudo R-squared (R2), Akaike Information Criteria (AIC), Bayesian Information Criteria (BIC), LL-null, and Log-Likelihood Ratio (LLR) are considered to investigate the performance of the proposed ABCDLR model. Furthermore, the proposed model has been applied for path planning in three different types of dense environments, and its performance is compared with four other competitive path planning approaches, such as, A*, Fuzzy Logic Controller(FLC), Vector Field Histogram(VFH) and ASGDLR.