Aiming at the problem that the Gaussian Process Motion Planner 2 (GPMP2) has limited ability to deal with the constraints of nonlinear inequalities and is prone to fall into local minimum in maps with complex obstaclesand thus may produce collisionsan improved SQPGPMP2 algorithm is proposed by combining with Sequential Quadratic Programming (SQP).Firstthe motion plan is regarded as trajectory optimizationand the initial trajectory state is obtained.Seconda collision cost function is introduced to represent the collision cost relationship between the robot and the obstacle.Finallythe SQP algorithm is used for iterative correction to ensure that the trajectory is collisionfree and dynamically reasonable.Simulation experiments show thatcompared with algorithms such as GPMP2the planning success rate of the algorithm on mazes of different sizes is improved by at least 20 percentage pointswhich proves that the algorithm is superior to them in dealing with the constraints and ensuring the planning efficiency.