To address the issues of slow convergence and low success rate in robot path planning associated with the DQN algorithm, this study introduces an enhanced version of the DQN algorithm named MDQN. This enhancement involves modifying the configuration of the state space to more accurately distinguish different states, resulting in a more realistic observation space. Additionally, by altering the reward function, the algorithm is capable of obtaining more reward values, thereby improving learning efficiency. Experimental evaluations conducted on grid maps demonstrate that the improved algorithm, when compared to the traditional DQN algorithm and other related approaches, offers advantages such as higher success rates, faster convergence, and shorter paths.