Artificial Intelligence-Based Sensorless Force Estimation for Smart Walkers: A Robust LSTM Disturbance Observer

Active smart walkers assist older adults with mobility. Traditional devices use physical force sensors to measure user intent. Physical sensors increase manufacturing costs and introduce mechanical fragility. Sensorless force estimation offers an alternative using disturbance observers to calculate external forces based on internal motor data. Traditional mathematical observers require exact physical parameters and velocity derivatives. Sensor noise and varying physical mass parameters cause estimation errors in these mathematical models. This paper presents a sensorless force estimation framework using a long short-term memory network. The network maps motor velocities, torques, and pitch angles to user forces. This approach removes the need for velocity derivatives and adapts to physical parameter changes. The study evaluates the framework through MATLAB simulations configured with the exact mechanical parameters of a physical prototype. The network filters sensor noise and tracks user force under a 6-kilogram mass mismatch. The integrated admittance control system separates user input from gravitational forces on inclined paths. The system executes differential turns and includes an automatic braking function that holds the walker stationary on a slope without user input. This study provides a framework for sensorless control in assistive mobility devices.