With regard to the dynamic obstacles current unmanned aerial vehicles encountered in practical applications, an integral suboptimal trajectory programming method was proposed. It tackled with multiple constraints simultaneously while guiding the unmanned aerial vehicle to execute autonomous avoidance maneuver. The kinetics of both unmanned aerial vehicle and dynamic obstacles were established with appropriate hypotheses. Then it was assumed that the unmanned aerial vehicle was faced with terminal constraints and control constraints in the whole duration. Meanwhile, the performance index was established as minimum control efforts. The initial trajectory was generated according to optimized model predictive static programming. Next, the slack variables were introduced to transform the inequality constraints arising from dynamic obstacle avoidance into equality constraints. In addition, sliding mode control theory was utilized to determine these slack variables' dynamics by designing the approaching law of sliding mode. Then the avoidance trajectory for single or multiple dynamic obstacles was developed by this combined method. At last, a further trajectory optimization was conducted by differential dynamic programming. Consequently, the integral problem was solved step by step and numerical simulations demonstrated that the integral method possessed high computational efficiency.
Person classification leveraging Convolutional Neural Network for obstacle avoidance via Unmanned Aerial Vehicles
