The invention provides a Slam method based on prior information. The method includes the steps of: 1) adding an initial pose node, extracting primary feature of initial radar observation, marking keyfeatures, adding a feature node to a map, and calculating the constraint and weight of the feature node and the initial pose node; 2) extending the map by means of observation, the prior information and the key features; 3) when the last frame of radar observation information is added to the map, performing global optimization to obtain a route and a feature map; 4) establishing a raster map of the environment on the basis of a Bayesian filter according to the route, which is obtained through the final optimization, of a movable robot and observation corresponding to every pose. The prior information is introduced through the expandable feature hierarchy, so that the method can adapt different environments and prior information in different types; besides the structural features, the key features are also considered, and the framework and size of the environment are further constrained by means of the extra constraint among the key features, thus reducing accumulated error.