The paper proposes a set of visual odometry and terrain reconstruction algorithms for quadruped robot in unstructured environments. Visual odometry can accurately measure the change value of the robot body's 6 degrees of freedom in 3D space. Terrain reconstruction can aid the robot to determine foot location around obstacles. With the visual odometry, the robot can remember the obstacle position even if the obstacles are out of vision. In order to simplify the process, two parts will also use the same sparse stereo visual image processing methods. The detail of the algorithms and the implementation process are introduced briefly. At last, the experiment results show that the algorithms are accurate and robust for quadruped robot navigation.
展开▼