A mobile robot usually works in dynamic environments with many uncertainties caused by either humans or various obstacles. Such uncertainties may cause unexpected error situations that often lead to navigation failure. Therefore, the robot should be able to recover from these unexpected error situations. This paper proposes an error recovery framework based on generalized stochastic Petri nets (GSPN). The approach can provide several advantages. The proposed framework can model various error situations occurring in real environments, thereby enabling a robot to recover from error situations autonomously. The modeling, analysis, and performance evaluation can be also carried out using the GSPN model. Experimental results show that the proposed error recovery framework is useful for dependable navigation of a mobile robot operating autonomously.