Autonomous navigation of unmanned aerial vehicles (UAVs) in unknown complex environments requires safe, fast and efficient path planning algorithms. Currently, the two-stage framework of “front-end search and back-end optimization” is widely adopted. However, existing research primarily focuses on path planning performance in high-density obstacle scenarios, lacking effective strategies for large-sized obstacles. Furthermore, the current two-stage framework suffers from issues such as path divergence and reduced flight speed. To address these limitations, this paper proposes a general path planning algorithm with soft constraints for UAVs in high-density obstacle scenarios and large-sized obstacle scenarios. The core of the algorithm involves guiding the UAV trajectory through the establishment of well-defined local target points. The front-end employs an expanded space observer for two observations, constructing a real-time safety region, and integrates flight state information to generate local target points using reinforcement learning. The back-end generates trajectories that allows UAVs to fly towards the local target points at higher speeds through an improved Soft Differential Constrained Minimum Snap (SDC-Minimum Snap) algorithm. For large-sized obstacles, a cost-function-based backtracking and circumvention mechanism is introduced to ensure reliable obstacle avoidance. Simulations and real-world experiments validate the generality and feasibility of the proposed algorithm in both scenarios.