Considering robotic visual inertial odometry(VIO) using filtering methods, a VIO algorithm is proposed in order to improve estimation accuracy. This algorithm uses cubature Kalman filter on matrix Lie group to realize it, which can accurately describe system uncertainty in rotation and reduce the linearization error of systems. The characters of the proposed algorithm are that: 1) the state is built by an high dimensional Lie group matrix and the definition of the additional operation for Lie group variant is proposed in cubature point sampling, which can extend the concept of cubature point, state mean and covariance from Euclidean space to manifold; 2)the state mean and covariance are propagated by cubature transformation, which avoids calculating complicated Jacobi matrixes and reduces the linearization error of the system. The performance of the proposed algorithm is tested in the EuRoc MAV dataset, and the results show the effectiveness of the proposed algorithm in improving estimation accuracy.