Navigation and positioning of UAVs in emergency situations during the construction of the subway and transport tunnels

Abstract

The article is devoted to ensuring reliable navigation and positioning of unmanned aerial vehicles in emergency scenarios typical for the construction of subways and transport tunnels, when satellite navigation is unavailable due to shielding, interference or suppression, and smoke and low illumination impair the operation of optical means. The limitations of existing approaches focused on partially controlled conditions are considered, and a multisensory architecture is proposed that combines an inertial navigation system for short-term autonomy, visual odometry to compensate for drift while maintaining visibility, UWB/Wi-Fi radio frequency beacons for global tethering in infrastructurally equipped areas, and lidar SLAM for operation in complete darkness. The key computational mechanism is adaptive correction based on the Kalman filter with a dynamic redistribution of sensor confidence through weights calculated as an exponential function of error variance estimates; it is shown that the sensitivity parameter γ sets a compromise between inertia and the sharp dominance of the «best» channel, and a temporarily variable setting γ is proposed for an accelerated response to sudden environmental changes (t), depending on the rate of dispersion growth. Numerical aspects of the implementation (exponential normalization, covariance symmetrization, Cholesky and SVD decompositions, iterative methods), as well as the extension of the model to the case of correlated noise through the introduction of a cross-сovariance matrix and asynchronous buffering measurements are discussed. Experimental testing in smoke simulation (density up to 0.7-0.8 units), in an underground tunnel with the placement of UWB beacons after 10 m and in an urban environment with GPS suppression demonstrates a decrease in error to 0.3-0.4 m in extreme conditions and an advantage of 3-10 times relative to monosensory solutions; additionally, a decrease in error growth with increasing length is shown a tunnel (up to 14 cm per 1000 m) compared to one INC. The limitations related to the computational load and requirements for on-board computers, as well as directions for further miniaturization and optimization, are indicated.