Abstract: Aiming at the key issues of insufficient coordination and proneness to falling into local optima in the obstacle avoidance path planning of unmanned underwater vehicle (UUV) formations, an Escape Algorithm (EA) that integrates crowd emergency evacuation behavior is proposed in this paper. The algorithm adopts an exploration-exploitation two-stage strategy, classifies agents into three categories (calm, follower, and panic) and designs differentiated position update mechanisms, and balances global exploration and local exploitation by combining an adaptive panic index and elite pool guidance. A hybrid reward function integrating navigation, formation maintenance, obstacle avoidance, and inter-UUV safety is constructed, which transforms multi-objective optimization into single-objective optimization to adapt to the requirements of underwater real-time decision-making. Validated by the CEC 2022 standard test functions, EA achieves an average ranking of only 1.58 under the 20-dimensional test configuration, with optimization accuracy, stability, and robustness all superior to the comparative algorithms. In the obstacle avoidance simulation of UUV formations, EA outperforms the comparative algorithms in comprehensive performance in terms of trajectory smoothness, formation maintenance stability, and obstacle avoidance efficiency. This algorithm effectively solves the key issues of insufficient coordination in path planning and proneness to falling into local optima in UUV formation obstacle avoidance, and provides a reliable path planning technical solution for cooperative obstacle avoidance of UUV formations.