Robotic navigation aid for visually impaired people using FLC-ORCA

Abstract

Robotic Navigation Aid (RNA) provides an option for visually impaired people for self-navigation and allow them to travel independently. However, current research does not take into consideration the varying characteristics and different states of obstacle which is present in the surrounding. Current methods of obstacle avoidance have been one-dimensional and limited in terms of representing the real-life surrounding when it comes to simulation and experimental conduct. Navigations are done either within a confined space of clear path, passing through static objects or between other mobile robots itself. Dynamic obstacles are presented within a controlled manner without differentiating class of objects which is intelligent and have the ability to avoid obstacles by itself. While separating static and dynamic objects are clear-cut and obvious, the same could not be said about differentiating moving object and intelligent entity with unpredictable movement. Moreover, the obstacle avoidance methods which attempt to solve the problem of collision within varying state of obstacles tend to assume that other intelligent entity follows the same obstacle avoidance protocol as itself. Furthermore, the presupposition that other intelligent entity bears equal liability of collision avoidance are too far from the circumstances faced in real life. Hence, the introduction of novel methodologies of CE-CBCE & FLC-ORCA attempts to fill in the void within two correlated and fundamental fields of object detection and obstacle avoidance. The proposed CE-CBCE feature extraction method detects human, cyclist and vehicles and separates it from other moving object to be classified as intelligent entity. Sequentially, FLC-ORCA resolves the issue of diversified position, velocity, acceleration, and density of obstacles and depicts the output of responsibility to avoid collision and predicts the obstacle character in the next cycle of scanning. The CE-CBCE feature extraction method recorded 97% detection and classification when tested using 1200 dataset consist of human, motorcyclist and vehicle point cloud data. The method is then tested with higher difficulty experiment, where it is required to recognize the class of the object and at the same time to determine the object’s pose detection. Impressively, the proposed CE-CBCE yields 82% of accuracy. Final experiments conducted within a living room environment with the presence of static obstacles, human as the intelligent entity and mobile robot TurtleBot 3 as rigid dynamic object moving without collision avoidance liability. Navigations are observed and its time of navigation, rerouting occurrence, change of heading angle, occurrence of stoppage and time of stoppage are recorded. Final results exhibit that the proposed FLC-ORCA method outperforms other state-of-the-art obstacle avoidance methods. Based on the experiment conducted, it can be inferred that the proposed FLC-ORCA allows navigation within static, dynamic and intelligent entity obstacles. It also prevails in avoiding collision within moving object with different collision avoidance protocol and different liability to circumvent obstruction. With the proposed method of CE-CBCE and FLC-ORCA, the risk of accidents and collision within the visually impaired people could be prevented. Therefore, allowing them to be independent while successfully achieving self-navigation in a safe, secure and unharmed manner.