A modified nonlinear fixed-interval smoothing filter is proposed for solving a problem with an SDINS (Strap-Down Inertial Navigation System) that uses tilt angle model as attitude error. Because the covariance of tilt angle is not equal to the covariance of the estimated attitude, the conventional smoothing filter is inadequate to solve such a problem. In the modified nonlinear smoothing filter, the backward filter is formulated by error states, so the outputs of the backward filter can be directly applied for the proposed smoothing algorithm, which combines the outputs of the forward filter and the backward filter. The smoothing filter is more efficient than the previous one because of its simple structure, and it can be widely applied for nonlinear smoothing filtering problems. The smoothing filter is applied to the off-line navigation of the PIG (Pipeline Inspection Gauge) using an SDINS, and its performance is examined by both computer simulation and experimentation. In the simulation, the proposed smoothing filter estimates position and attitude of the PIG well, and the efficiency of the proposed method is confirmed. In the experimentation, the smoothing filter is applied to solve the navigation problem using real data obtained from the PIG developed by Korea Gas Company and Seoul National University. To verify the performance of the smoothing filter, five check points on the pipeline are selected and precisely measured by the CDGPS. The difference between navigation results and the measured positions is consistently less than 1 m. From the simulation and experimentation results, the performance and feasibility of the modified nonlinear smoothing filter is verified.