Vector field-based motion planning for autonomous exploration in mining environments

Abstract Underground mining environments are renowned for their vast scale, confined spaces, and limited visibility. This paper addresses the issue of exploration through exceptionally narrow tunnels by framing it as a problem of autonomous navigation in constrained environments while ensuring collision avoidance. To tackle this challenge, a composite algorithm consisting of two phases: reference path generation and collision-free path following, is introduced. In the initial phase, a short window of underground environmental data acquired from an onboard sensor is processed to extract the tunnel’s skeletal structure within the local window. Subsequently, smoothing techniques are applied to refine this tunnel skeleton, generating a reference path for the robot to track. In the second phase focusing on path following, the concept of zero-level set is introduced to represent both the tunnel walls and any encountered obstacles that might obstruct the reference path. These representations are approximated using suitably smooth radial basis functions. By leveraging the principle of vector field, the robot is guided along the reference path while simultaneously avoiding collisions with arbitrarily shaped obstacles. A significant advantage of this approach lies in its generality and adaptability. Unlike existing algorithms, the desired reference path is straightforwardly derived from the tunnel skeleton. The only requirement for the shapes of the reference curve and sensed obstacles is that they should be homeomorphic to a real line or a circle. Moreover, the adopted vector field algorithm ensures both path following and collision avoidance without reliance on specific geometric constraints. The proposed strategy has been implemented and rigorously tested under both synthetic and real-world scenarios, with results demonstrating the effectiveness of the system and the advantages of automation in this particular application.