The basic fast random number (RRT) algorithm has some shortcomings, such as randomness, lack of directionality, long path generation time and low efficiency. On this basis, an algorithm based on improved RRT path planning is proposed. Firstly, the target bias strategy is added to the random sampling expansion, so that the growth direction of the random tree tends to the target point, and the target orientation of the sampling is enhanced; Then the node rejection strategy and variable step size strategy are adopted to enhance the convergence speed of the algorithm and the obstacle avoidance effect of the manipulator; Finally, on the basis of improving the convergence conditions, the greedy algorithm is used to remove some redundant nodes, simplify the obstacle avoidance path, and the cubic B-spline curve is used to smooth the generated path, further optimize the path length, and make the generated final path smoother. The improved RRT algorithm, the standard RRT algorithm and the RRT-Connect algorithm are simulated in different environment maps. The comparison of the experimental results shows that the improved RRT algorithm has been improved in planning time, node number, path length and other aspects, and verifies the feasibility and superiority of the improved scheme.