Abstract: The intelligent control of rock drilling rig boom's hole-seeking process is crucial for enhancing the accuracy and efficiency of rock drilling operations. Inverse kinematics solution (IKS) is the core of achieving precise and rapid hole-seeking control of the boom. Existing analytical and numerical techniques are inadequate in fulfilling the accuracy and time efficiency demands for IKS in complex drilling boom scenarios. Conventional neural network (NN) approaches, heavily dependent on labeled data, often fall short in solution reliability. To overcome these challenges, this study introduces an unsupervised learning-based NN method for IKS, with an emphasis on safety constraints. This novel method diverges from traditional approaches by not relying on labeled data. It employs the desired end position of the drilling boom as the input for the network, focusing on minimizing the discrepancy between the actual and intended end positions. A critical innovation of this study is the integration of a safety collision penalty into the solution's objective function, ensuring the network's output for joint positions adheres to specific environmental limitations. Empirical evidence demonstrates that this unsupervised learning-based method achieves a mean hole-seeking error ranging from 5 to 7 mm in IKS, a significant improvement of about 238.46% over supervised learning methods. Moreover, the introduction of safety constraints has successfully reduced the collision rate in IKS solutions by 92.15%, without any sacrifice in the accuracy of the solutions.