Path Planning under Goal Position Uncertainty in Loading Scenarios Based on Convolutional Neural Network

In current truck-excavator collaboration operation, autonomous mining trucks are required to plan a path from their starting point to a parking position near the excavator. However, the parking position for path planning remains uncertain, requiring the trucks to wait at the starting point until the excavator operator provides the parking heading, which significantly hampers production efficiency. Existing path planning methods rely on an implicit assumption of precise target position knowledge, rendering them ineffective in handling scenarios with uncertain target positions. This paper proposes a novel path planning method to address this issue, employing a neural network to predict the parking angle as a substitute for manual input, and thereby designing a method to calculate the parking position. Subsequently, the variable-step Hybrid A* algorithm is utilized to generate a collision-free path from the starting point to the parking position. Experimental results demonstrate that the neural network can accurately predict parking heading, effectively replacing manually provided parking heading. The proposed method allows autonomous trucks to plan their paths in advance during the excavator’s preparatory tasks, thereby improving overall production efficiency.