Abstract: To address the issues of insufficient consideration of agent heterogeneity and the integration of human experience and cognition in path planning research, this paper proposes a human-machine collaborative decision-making heterogeneous multi-agent path planning method in a mixed reality scenario. First, a heterogeneous multi-agent deep reinforcement learning method is introduced, based on danger-guided points and RVO (reciprocal velocity obstacles) for local collision avoidance. This method provides global guidance and local instructions according to agent heterogeneity, with a reward function tailored for heterogeneous agents to address the sparse reward problem. Then, during the virtual-physical agent interaction process in mixed reality, human experience is incorporated to adjust the paths generated by intelligent methods, enhancing human guidance and enabling human-machine collaborative decision-making in path planning to compensate for the limitations of intelligent methods. Experimental results in 2D, 3D, and mixed reality scenarios demonstrate that the proposed method not only applies to heterogeneous multi-agent path planning but also achieves collaborative decision-making planning in mixed reality. It outperforms comparative methods in success rate, convergence, path length, and number of turns.