A real-time predictive postural control system with temperature feedback

Balanced posture is essential in sports training, rehabilitation therapy, and robotic control. The application of biofeedback technology has significantly improved postural stability, particularly in individuals with sensory disorders. In practical applications, thermal biofeedback is regarded as an optimal method for enhancing posture control. However, conventional systems frequently encounter challenges with slow temperature adjustments, resulting in delayed responses. Thus, enhancing the responsiveness of these temperature control mechanisms is critical for achieving better real-time performance. In this study, we designed a system incorporating smart sensors to support balance correction and postural stability. The designed system employs inertial sensors to measure body tilt angles and a wearable temperature control module for biofeedback. Moreover, we proposed a mathematical method to improve the real-time biofeedback with thermal tactile feedback, specifically targeting the issue of poor real-time temperature regulation. An Long Short-Term Memory (LSTM) neural network with a sliding window method is incorporated to predict the posture patterns of the next state. In order to optimize the bidirectional LSTM training process, cross-validation is utilized to assess model performance. This predictive strategy accelerates thermal perception and facilitates immediate balance restoration. Experiments were conducted to evaluate the system's reliability and effectiveness in balance correction. Compared to the typical thermal feedback and angle change, the results demonstrate that the LSTM network accurately predicts posture changes based on angular and acceleration data, enabling more timely temperature adjustments and significantly enhancing balance ability, as validated through Romberg standing tilt test evaluations.