<p>Household service robots face significant challenges in obstacle avoidance and environmental adaptability when performing grasping tasks in cluttered daily life scenarios. Existing approaches often suffer from inaccurate environment modelling and slow convergence in high-dimensional joint space planning, which limits their effectiveness in complex household environments. To address these issues, this paper proposes an obstacle avoidance path planning method for robot grasping based on point cloud environment modelling. The method consists of two main modules: an Environment Modelling Module and an Obstacle Avoidance Path Planning Module. In the Environment Modelling Module, an octree-based environment representation combined with point cloud conditional filtering is employed to efficiently obtain accurate spatial occupancy information while suppressing self-occupancy caused by the robotic arm. In the Obstacle Avoidance Path Planning Module, a Decoupled Pose Planning Strategy incorporating dimensionality reduction is introduced to accelerate optimization convergence in the robotic arm joint space. Furthermore, an efficient collision detection method is developed to enhance the obstacle avoidance capability of the robotic arm during planning. Simulation results demonstrate that the proposed strategy achieves 31.1–67.6% lower planning time and generates safer collision-free paths compared with baseline methods. Real-world grasping experiments in replicated household scenarios further validate that the proposed method can reliably produce feasible and safe obstacle avoidance paths.</p>