UAV Navigation in Tunnels with 2D tilted LiDARs

TL;DR

Problem: UAVs operating in GNSS-denied tunnels must navigate at high speed despite uneven illumination and scarce features, making traditional localization hard. Approach: a perception-steer pipeline using two tilted 2D LiDARs to estimate yaw relative to the tunnel axis via a CNN, plus a geometric module (pole of inaccessibility) to locate the safest position and guide lateral speeds; roll and pitch are compensated with inclinometer data. Contributions: a proof-of-concept showing feasible high-speed navigation without full localization, validated in Gazebo simulations across bent tunnels up to $6\\ \mathrm{m/s}$ with controlled heading error. Impact: enables robust tunnel navigation using lightweight sensors and end-to-end processing, reducing reliance on maps, odometry, and rich visual features.

Abstract

Navigation of UAVs in challenging environments like tunnels or mines, where it is not possible to use GNSS methods to self-localize, illumination may be uneven or nonexistent, and wall features are likely to be scarce, is a complex task, especially if the navigation has to be done at high speed. In this paper we propose a novel proof-of-concept navigation technique for UAVs based on the use of LiDAR information through the joint use of geometric and machine-learning algorithms. The perceived information is processed by a deep neural network to establish the yaw of the UAV with respect to the tunnel's longitudinal axis, in order to adjust the direction of navigation. Additionally, a geometric method is used to compute the safest location inside the tunnel (i.e. the one that maximizes the distance to the closest obstacle). This information proves to be sufficient for simple yet effective navigation in straight and curved tunnels.

Figures (15)

• Figure 1: Dynamic reference for UAV navigation. The yaw ($\psi_d$) of the UAV and its $y$ coordinate is expressed with respect to the longitudinal axis. The angles $\gamma$ and $\psi_w$, useful for evaluating the performance, can be obtained from the simulated environment.
• Figure 2: UAV with LiDARs on top and below.
• Figure 3: LiDAR readings deformation in a cylindrical pipe. Vertical LiDARs with null yaw (a). Vertical LiDARs with not null yaw (clockwise and counterclockwise) (b). Tilted LiDARs with non-null yaw (clockwise) (c). Tilted LiDARs with non-null yaw (counterclockwise) (d).
• Figure 4: UAV with tilted LiDARs.
• Figure 5: Diagram of the CNN architecture.