Quadrotor drone manipulation using neuro fuzzy algorithm with non-invasive brain-computer interface

• Main Author: Chu, Timothy Scott C.
• Format: text
• Language: English
• Published: Animo Repository 2020
• Subjects: Adaptive control systems; Neuroadaptive systems (Bioengineering); Quadrotor helicopters; Mechanical Engineering
• Online Access: https://animorepository.dlsu.edu.ph/etd_masteral/5978; https://animorepository.dlsu.edu.ph/context/etd_masteral/article/12908/viewcontent/CHU_TimothyScott_11886072_Quadrotor_Drone_Manipulation_using_a_Neuro_Fuzzy_Algorithm_with_Non_Invasive_Brain_Computer_Interface_Redacted.pdf
• Institution: De La Salle University

Brain-Computer Interface or BCI machines are devices that obtain signals generated from the brain and are then manipulated to suit various applications, such as interfacing with robotics. These researches usually utilize different Machine Learning algorithms. Most studies used LDA, Decision Trees, ANN, and CNN algorithms and returned relatively acceptable average accuracy percentages. The challenge, therefore, is to develop or employ a Machine Learning algorithm that offers a good balance between computational resource requirements and accuracy. This study aimed to analyze the performance of a Neuro-Fuzzy algorithm, specifically the Adaptive-Network-Fuzzy-Inference System (ANFIS), to classify EEG signals retrieved by the Emotiv INSIGHT, and its viability to be implemented to a quadrotor system. The research started by designing a system framework, which reflected how key processes and components collectively function to meet the main goal of maneuvering the quadcopter with a BCI control system; this is followed by an analysis of the performance of the ANFIS algorithm in conjunction with the SVM algorithm, where the latter algorithm served as a reference. Hardware limitations prompted the researcher to shift from using the Steady-State Visual Evoked Potential to Facial and Eye Gestures to generate EEG signals. Raw and Filtered EEG datasets were fed to both algorithms for simulation experiments and test flight experiments. Parameters such as accuracy, training time, and prediction time were considered in the evaluation of the algorithms. Results showed that the ANFIS is capable of maneuvering the quadcopter with an accuracy rate of 78.67% with the filtered dataset. A flight path experiment was also conducted, where this test showed how well the BCI-UAV system maneuvered the drone on a specified flight path. It was also found that the ANFIS took significant amounts of computational resources, such as RAM and training time; however, it did offer more precise results. At the end of the paper, the researcher presented recommendations that can improve the system employed by the study, as well as recommending processes that can extend this research to use mental commands to maneuver a drone.