A Q-learning based path planning method with adaptive steering constraints for USV is proposed in this paper. Firstly, the classic Q-learning algorithm in reinforcement learning is applied to the path planning for the USV, and the penalty functions for steering constraints applicable to the properties of the USV are defined. Then, a Q-learning method with multi-step foresight ability is proposed to address the shortcomings of the classical reinforcement learning algorithm. On the basis of the original definition of constraints, the penalty factor is adaptively adjusted, from which a modified Q-learning based fusion algorithm with adaptive steering constraints is established. The simulation results in the rasterized electronic chart environment demonstrate the effectiveness of this control method in reducing the steering of USV and its strong predictive ability in path planning.