An improved method for odometry estimation based on EKF and Temporal Convolutional Network

Xiaokun Leng, Songhao Piao, Song Wang, Lin Chang, Zheng Zhu

Abstract: Compared with wheeled robots, the walking direction and distance of biped robots are difficult to accurately control, and the body of the robot will periodically shake during walking. During the walking of the biped robot, the movement of the torso drives the head camera to shake, which causes great interference to the slam algorithm, and leads to deviations in the vision-based mileage estimation method. This paper mainly combines the vision positioning method and the sensor data on the biped robot and estimates the position information of the robot through data fusion. This method makes full use of the sensor data on the robot to improve the estimation accuracy of the odometer.