An appropriate stepsize is required to be set up when using rapidly exploringrandom tree (RRT) to perform path planning of a robot, 〖JP2〗which needs user to proceed debugging the program and it’sgenerally timeconsuming, also a fixed stepsize in RRT always resulting in invalid collisiontest. Aiming at solving the above problems existing in RRT, a selfadaptive stepsize RRT was proposed. The matrix operator norm induced from configuration space norm and work space norm was founded based on Jacobi matrix and the norm inequality of configuration space and work space was established, by the means of which the displacement of robot caused by each stepsize in configuration space was limited in allowed magnitude which validated collision test. In order to coordinate dualrobot, passive growing of random tree algorithm was put forward. The algorithm can control the growth of random tree of dualrobot in different configuration spaces, and then the motion of dualrobot was coordinated to ensure generating cooperation path in work space. Numerical experiment indicated that the selfadaptive stepsize RRT can bound the displacement of each step within the value set up at beginning of algorithm which guaranteed the effectiveness of collision test. Compared with standard fixed stepsize RRT, selfadaptive stepsize RRT omitted the process of determining stepsize only needed to set maximum value of stepsize in work space which improved the efficiency of path planning. The algorithm proposed can provide a new perspective on the path planning of dualarm robot.