Fuzzy Q-Learning-Based Adaptive Admittance Control Method, Device Performing the Same, and Computer Program

Adaptive Admittance Control Technology for Wearable Robots Using Fuzzy Q-Learning

This technology utilizes a fuzzy Q-learning algorithm to adaptively adjust the admittance parameters of wearable robots. It monitors human-robot collaboration status in real-time using low-pass and high-pass filters, using this data along with interaction forces and load weight as inputs for a fuzzy control model to calculate optimal control values.

Setting fixed admittance gains created a trade-off between sensitivity and safety, while existing DFT-based frequency analysis methods struggled with observation delays, making it difficult to recognize collaboration status or respond to impacts within 0.5 seconds.

This technology achieves low-latency collaboration recognition using a second-order IIR Butterworth filter and optimizes parameters in real-time by applying fuzzy Q-learning to expert knowledge-based fuzzy rule initial values. It serves as an innovative solution for power-assist suits and logistics wearable robots by automatically balancing sensitivity and safety based on the situation.

‍

‍

Key Features:

• A step of acquiring interaction force and load weight in wearable robots where humans and robots are in constant contact
• A step of acquiring human-robot collaboration status values based on interaction force using low-pass and high-pass filters
• An adjustment model that takes collaboration status values, interaction forces, and load weights as input data to output admittance parameter values
• A configuration that acquires admittance parameter control values using an adjustment model based on fuzzy Q-learning