<p>In complex environments with multiple obstacles, it is crucial to plan trajectories that avoid collisions between both the end effector and the joints of pneumatic soft manipulators and surrounding obstacles, thereby placing higher demands on collision detection and obstacle avoidance algorithms. Currently, the collision detection for soft manipulators is mostly based on the configuration space, making it difficult to reflect the actual collision relationship between obstacles and joints directly, especially when the configuration space parameters do not change significantly. Moreover, research on obstacle avoidance algorithms tailored for pneumatic soft manipulators is still relatively scarce. To address these issues, this paper presents a trajectory planning method for a pneumatic soft manipulator with four chambers, focusing on obstacle avoidance. Firstly, a collision detection method is developed within the task space, which accounts for the physical structure characteristics and real-time deformation behavior of the pneumatic soft manipulator. Subsequently, an enhanced rapidly-exploring random tree (RRT) algorithm is developed to efficiently generate a collision-free trajectory with reduced path redundancy and improved smoothness. Finally, the effectiveness of the proposed obstacle avoidance trajectory planning method is validated through simulations and experiments.</p>