Abstract:
Robot’s ability to perform its tasks safely, securely and efficiently is heavily dependent on understanding of its operational environment. Understanding robot’s environment therefore precedes devising approaches to accomplishing set tasks for a robot. In an unstructured environment it is very challenging for a robot to navigate because it must be capable of identifying and adapting to changes. In essence, an unstructured environment is chaotic and unpredictable to a robot navigation. In this paper, machine learning techniques were used to train robot for optimum performance in recognition of an unstructured environment. Feature selection methods were used to extract relevant features from a standard dataset generated by a self-driving car robot at the Oakwood University. After feature selection, the environment were classified using Naive Bayes, K-Nearest Neighbour, Multilayer perceptron and Support Vector Machine (SVM) machine learning techniques for recognition of the resulting dataset at similar scenarios. The data analysis and interpretation shows SVM to be 0.66% accuracy, KNM 0.65%, Decision Tree 0.66% and MLP 0.65% accuracy. Hence, SVM outperform others in the classifier. The result shows that robot can understand its environment when there are changes in the environment or when there are different scenarios in the environment.
