Method for predicting UAV trajectories and evasion for industrial autonomous missions in a dynamic environment

This paper addresses the problem of predicting the trajectory and anticipatory evasion of an unmanned aerial vehicle (UAV) in industrial autonomous missions in a dynamic environment, subject to noise in navigation measurements and energy constraints, which pose risks of delayed or excessive maneuvering and increased deviation from the route. Objective. To develop and verify, using a simulation model, a method that ensures prediction-based obstacle avoidance with control of deviation from the reference trajectory and maneuvering energy consumption. Tasks. Formulate the architecture of the avoidance system; develop a predictor of future coordinates based on a recurrent neural network with long-term short-term memory; determine a method for assessing collision risk using a safety zone; implement a trajectory correction algorithm taking into account the “safety–deviation–energy consumption” trade-off; perform a comparative evaluation with baseline methods. Methods. Coordinate predictions are constructed based on time sequences of coordinates and motion parameters; collision risk is assessed by analyzing the intersection of the predicted trajectory with obstacle safety zones; trajectory correction is formalized as an optimization problem to select a maneuver that minimizes the total tracking error and the proximity penalty. The effectiveness was verified in a Python environment on standard trajectories (straight, circular, and polygonal) by comparison with pure tracking, line-of-sight, vector field, and nonlinear stabilization methods. Results. The proposed approach achieved the smallest mean deviation from the trajectory (14.95 m), the lowest maneuver energy consumption (72 conventional units), the highest tracking success rate (86.08%), and the highest overall productivity coefficient (0.494) among the algorithms considered; a trade-off was observed regarding the minimum distance to obstacles. Conclusions. The prediction-oriented evasion method improves the overall navigation efficiency in industrial mission models; further research involves field validation on real platforms and optimization of the predictor’s computational costs.