Abstract: Real time navigation of mobile robot in globally unknown environment is studied. A new method for mobile robot's navigation is presented with the combination of describing environment based on grids and path planning based on rolling windows. The environment information detected by sonar is represented in the form of probability based on grids, and is fused by evidential reasoning with uncertainty based on Dempster-Shafer theory to describe the local environment around the mobile robot. And then, path planning based on rolling windows is implemented to realize real time navigation of mobile robot. Simulation and experiment results show that the method is effective.