Odometry Estimation Utilizing 6-DOF Force Sensors and IMU for Legged Robot

摘要

Odometry estimation is the problem of estimating platform pose utilizing different types of sensors. Visual odometry calculates the translation and rotation of two frames by extracting and matching feature points. However, it cannot function properly in some situations, such as dark environment or scenes with repeated textures. In rescuing scenario with poor visibility, fire-fighting robots and nuclear power plant rescue robots can not localize itself by solely depending on visual information. A real-time high precision odometry estimation method based on 6-DOF force sensors and IMU for legged robot, which can overcome the above limitations, is proposed in this paper. The odometry estimation method is realized by real-time robot posture estimation through 6-DOF force sensors mounted on the legs of legged robot. We combine particle filter and Quasi RNN network to build particle filter net. Through the fusion of 6-DOF force sensors and IMU using particle filter net, real-time odometry estimation with high precision is achieved. Compared with the state of art, our method has advantages of higher accuracy, faster computation speed and broader application scenarios for legged robot.