Abstract

Aiming at the engineering problems of composite roll, pitch, and yaw pose errors of the roadheader body induced by floor undulation and geological variation, as well as insufficient anchoring accuracy caused by the incapability of traditional 1–3-degree-of-freedom (DOF) drilling–anchoring arms in dynamic error compensation during coal mine roadway excavation and bolting, this paper proposes an inverse kinematics solving method for a 5-DOF drilling–anchoring arm based on an improved Deep Deterministic Policy Gradient (DDPG) algorithm. Firstly, the modified Denavit–Hartenberg (MDH) approach is adopted to establish a full-link kinematic model incorporating body pose errors, where the drill rod length, mounting offset, and world coordinate transformation are fully considered. Secondly, an Actor–Critic dual-network architecture tailored for drilling and anchoring tasks is constructed with an 11-dimensional state space and a 5-dimensional action space. The coupling optimization between body pose errors and joint adjustments is realized by designing a hierarchical gradient reward function, a dynamic noise decay exploration strategy, and an optimal state restart mechanism. Finally, 1000 episodes of training and verification are carried out on a Python 3.10 simulation platform. The simulation results reveal that the average end-effector position error of the improved algorithm reaches 0.71 mm, and the deflection angle toward the roof is less than 1°, which outperforms the specified industrial standard. The proposed method realizes real-time compensation for dynamic body pose errors and provides crucial technical support for intelligent excavation and anchoring in underground coal mines.

IPC Classification

Keywords