Human Velocity Estimation Using Kalman Filter and Least Squares With Adjustable Window Sizes for Mobile Robots

For autonomous mobile robots to work safely in human-coexistent environments, human-velocity estimation is essential. However, the human body periodically fluctuates to the front, rear, right, and left while walking. Also, a significant estimation error occurs due to the vibration of sensors installed in the robot. Quick trajectory adjustment requires high-accuracy and low-latency estimation, but these are in a trade-off relationship. We thus propose a human velocity estimation system (VES) using the Kalman filter (KF) and least squares (LS) with adjustable window size (AWS) to control the accuracy and latency. The VES adjusts two window sizes to calculate a system noise distribution for KF and a velocity vector for LS using a newly proposed cost function, including accuracy (direction and magnitude) and latency (time delay) costs. To select window sizes suitable for walking trajectories and individual gaits, we collected human walking data, calculated the three costs, and selected the window sizes with the minimum cost. The results of experiments using a laser range finder installed on a mobile robot indicate that the cost function could reveal window sizes to increase accuracy or reduce latency depending on walking trajectories and individual gaits, and the VES with AWS could enhance the performance of estimating human walking velocity for mobile robots.