In response to the core issues of the bidirectional rapidly-exploring random tree (RRT-Connect) algorithm in robotic arm path planning, such as low search efficiency, poor path quality, and weak adaptability to complex environments, this paper proposes a robotic arm path planning algorithm called H-RRT-C, which integrates hierarchical heuristic guidance and reinforcement learning. The method constructs a multi-strategy collaborative optimization system: The upper layer uses an improved A* algorithm to generate a global coarse-grained path skeleton, and adopts an adaptive weight mechanism to guide the bidirectional search tree to preferentially sample key nodes, effectively reducing the blindness of random exploration; The lower layer introduces a Dijkstra local search mechanism, which dynamically adjusts the search range according to the distribution density of obstacles to achieve fine-grained processing of local paths. At the same time, a double $ Q $-network reinforcement learning strategy is introduced, and a multi-objective reward function including path length, node distribution diversity, and obstacle avoidance safety is designed to realize intelligent decision-making for the expansion direction. Finally, Matlab simulation experiments verify the path planning effectiveness of the proposed algorithm in various complex scenarios, and tests conducted on a ROS platform and a physical robotic arm validate its engineering practicality.