Uncertainty-Constrained Motion Planning forTerrain-Aided Navigation

Sammanfattning

Aerial vehicles operating in GNSS-denied environments must rely on alternative navigation methods. Inertial Navigation Systems (INS) can propagate the vehicle state without external signals, but the estimate drifts over time due to the accumulation of errors. Position measurements from a Global Navigation Satellite System (GNSS), such as GPS, are typically used to correct this drift. In GNSS-denied environments, however, GNSS may be unavailable or unreliable. Terrain-Aided Navigation (TAN) is a positioning method that does not rely on GNSS. Instead, TAN compares terrain measurements with an elevation map to estimate the vehicle position and correct INS drift. However, the performance of TAN depends strongly on the terrain along the flown path. This creates a coupling between motion planning and navigation performance: a short path through weakly informative terrain may be difficult to execute robustly, while a longer path through more informative terrain may improve localization. This thesis investigates uncertainty-constrained motion planning for a fixed-wing-inspired aerial vehicle that uses TAN for navigation. An RRT*-based motion planner is modified to account for predicted navigation uncertainty by propagating the parametric Cramér--Rao lower bound (ParCRLB) along candidate paths. The planner minimizes travel time while enforcing deterministic constraints based on the propagated ParCRLB, including a path-wise trace constraint and a ParCRLB-based no-flight-zone safety margin. A multi-track propagation method is also introduced to make the uncertainty evaluation less dependent on a single nominal path. The planner is evaluated in closed-loop Monte Carlo simulation. The results show that constraining the ParCRLB can improve mission success, path-following performance, and estimation performance, but at the cost of longer planned paths. In the trace-threshold study, low maximum allowed ParCRLB trace values generally lead to higher mission success rates and lower estimation errors, while high trace thresholds allow shorter but less robust paths. The planner-mode comparison shows that enforcing low predicted uncertainty along the entire path gives the most robust behavior, while using only a ParCRLB-based safety margin around no-flight zones can be sufficient in less safety-critical scenarios. Overall, the results indicate that the ParCRLB is useful as a planning-stage proxy for navigation robustness, although it should not be interpreted as an exact prediction of the closed-loop estimation error.