Path planning is the core challenge for efficient and safe operation of building material handling robots in intelligent construction, especially in dynamic and complex working conditions, traditional methods often face problems such as path oscillations, obstacle avoidance failures and global-local decision mismatch. An A* and proximal policy optimization (PPO) collaborative optimisation framework is proposed to address this problem. The specific approaches are as follows: 1) An A* algorithm is improved by using eight-neighbourhood extension with Chebyshev distance to construct a gridded global topology; 2) A six-dimensional observation vector-driven dynamic reward function is designed, which integrates path tracking rewards, collision penalties, and step-size constraints; 3) A feature-level parameter-sharing mechanism is established, and A* path features are embed into the proximal policy optimization network using a dynamic window approach (DWA) to achieve global cost estimation and local obstacle avoidance decision synchronous optimisation. Simulation in four types of typical site environments shows that, compared with the RRT*-APF and traditional A* algorithms, the proposed method improves the success rate of the path in dynamic obstacle scenarios by 42.7% (the traditional algorithm fails), reduces the planning time by 55.8%, and successfully avoids obstacles 98 times when U-shaped concave obstacles are coupled with dynamic interference. Through progressive waypoint verification and two-layer optimisation architecture, the compatibility problem between topology maintenance and real-time obstacle avoidance is solved, and the path oscillation and obstacle avoidance delay problem of the construction robot is solved.