For a self-reconfigurable robot,how to metamorphose to adapt itself to environment is a difficultproblem.To solve this problem,a new relative orientation model which describes modules and their sur-rounding grids was given,a module motion rules database which enables the robot to avoid obstacles wasestablished,and finally a three-layer planner based on dynamic meta-modules was developed.The first-layer planner designates the category of each module in robot by evaluation functions and picks out themodules in dynamic recta-modules.The second-layer planner plans the dynamic recta-module path ac-cording to output parameters of the first-layer planner.The third-layer planner plans the motion of themodules in dynamic meta-module using topology variation oriented methods.To validate the efficiency ofthe three-layer planner,two simulations were given.One is the simulation of a single dynamic meta-mod-ule,the other is the simulation of planning with an initial configuration composed of 8 modules in compli-cated environment.Results show that the methods can make robot with any initial configuration movethrough metamorphosis in complicated environment efficiently.