Arithmetic Gradient Optimized Deep Residual Path Planning Model for 3D Environment in Unmannered Aerial Vehicles

Unmanned Aerial Vehicles (UAVs) are utilized in various applications that necessitate effective path planning strategies. Nevertheless, several algorithms developed recently may not be practical or efficient, especially when dealing with complex, three-dimensional (3D) flight environments. This paper considers real-time path planning based on global and local environmental data using a deep learning approach. For learning the behavior of the UAV state, the obstacle and distance information is trained using the Cascaded Residual Dense Block Network (CRDBN) model. CRDBN offers a solution that preserves both linear and non-linear correlations between state and behavior. Moreover, the hyperparameters of CRDBN are optimized using the Arithmetic Gradient Optimization (AGO) algorithm that ensures precise path planning. AGO makes the network more scalable in the direction of ideal solutions. The tests are carried out using the MATLAB software, and the effectiveness is assessed using metrics related to deep learning as well as efficiency, energy, and accuracy. The proposed method uses 866.73 J of energy while improving the path planning accuracy to 98.32.