Model predictive control based on incremental occupancy grid map for underground utility tunnels inspection UAV

In response to the challenges of autonomous perception, localization, and control of unmanned aerial vehicles in global navigation satellite system (GNSS)-denied underground utility tunnels, this paper proposes a model predictive control method based on an incremental occupancy grid. A lightweight LiDAR is used for environmental perception to generate point clouds, which are sparsely represented and hierarchically processed via an incremental occupancy grid mapping algorithm. This approach reduces computational and storage overhead while minimizing the discretization loss of environmental features. A data-driven hierarchical trajectory planning algorithm is then designed: at the local map level, Gaussian convolution kernels are applied to the occupancy matrix to generate real-time motion directional paths; at the global map level, non-uniform rational B-spline-based global directional constraints are incorporated to optimize and produce smooth, executable trajectories. Building on this framework, occupancy grid state constraints, waypoint constraints, and terminal stability constraints are introduced, enabling the MPC to achieve rapid and stable trajectory tracking in complex and confined environments. Finally, real-world flight experiments verify that the proposed method efficiently performs environmental perception, real-time planning, and precise tracking without prior maps, effectively solving the autonomous navigation and control problem of quadrotor UAVs in GNSS-denied scenarios and providing a reliable solution for autonomous inspection of urban underground utility tunnels.