Diseño y comparación de modelos de reconocimiento de gestos de la mano combinando aprendizaje supervisado y aprendizaje por refuerzo : Diseño de un modelo de reconocimiento de 5 gestos de la mano con función en tiempo real mediante el uso de Double Deep Q-Network (DDQN).

Abstract

Hand Gesture Recognition (HGR) using Electromyography (EMG) signals is a crucial component of Human-Computer Interaction, enabling natural communication between humans and technology in various applications such as prosthetics, robotics, and rehabilitation devices. While supervised learning techniques like Convolutional Neural Networks (CNNs) have been extensively explored in HGR systems and have achieved high levels of accuracy, reinforcement learning (RL) stands out for its ability to learn from interaction and solve sequential decision problems. Algorithms such as Q-Learning and Deep Q-Network (DQN) have shown promising results in HGR and EMG. This research aims to compare the performance of reinforcement learning techniques DQN and DDQN in HGR using the EMG-EPN-612 dataset, evaluating the accuracy of the models in both recognition and classification. Key contributions include the design of a reward function for this context and the use of a pre-existing convolutional neural network for fine-tuning. The results show that models with fine-tuning, alongside DQN and DDQN, exhibited similar values in the proposed metrics. A 3.61% improvement in recognition accuracy was observed for the DQN model and a 3.45% improvement for the DDQN model compared to the baseline model. However, no significant improvements were found in classification accuracy. In conclusion, it was found that DQN outperformed DDQN in this scenario.