[4K3-GS-10] AI application: Movement

[4K3-GS-10-03] Application of Machine Learning to Self-Position Estimation of Bulldozers Using Internal Sensors

〇Hikaru Sawafuji¹, Ryota Ozaki², Takuto Motomura², Toyohisa Matsuda², Masanori Tojima², Kento Uchida¹, Shinichi Shirakawa¹ (1. Yokohama National University, 2. Komatsu Ltd.)

Keywords:self-position estimation, machine learning, bulldozer, internal sensors, extended Kalman filter(EKF)

The self-positioning system of mining bulldozers uses the global navigation satellite system (GNSS), but GNSS sometimes does not respond at some mining sites. We proposes a self-positioning system of mining bulldozers without GNSS. Our system estimates the self-position in two steps: the estimation of velocity in local coordinates using the machine learning model and the self-positioning estimation in global coordinates. The velocity estimation system utilizes information from the internal sensor, including the inertial measurement unit (IMU). The global position is estimated by the extended Kalman filter (EKF) with estimated local velocity. The proposed system outperformed the competitive method without the machine-learning model.