In the current higher education system, physical education is faced with the development needs of integrating traditional teaching mode with modern technology. This paper proposes a multi-intelligent dynamic design method for college cheerleading that integrates the shortest path algorithm and reinforcement learning. The study adopts the MAPPO algorithm improved based on noise assistance to construct a multi-intelligence collaborative path planning model, and conducts training in a simulation environment with a side length of 10 m. The maximum speed of the intelligences is set to 0.90 m/s, and the learning rate is 0.003. The effect of the algorithm is verified through a comparative experiment with 102 students from a teacher training college in a certain city. The experimental results show that the students in the experimental class significantly improved in the four dimensions of movement technology, emotional expression, formation transformation and overall presentation, in which the emotional expression improved the most to reach 15.81 points, and the movement technology posttest score of 86.24 points was significantly higher than that of the control class of 78.23 points. The improved algorithm performs well in intelligent body collision rate control, with the CBRS value stably controlled near 1 and the CBRO value maintained in the range of 0.5-2.0. The study proves the effectiveness of reinforcement learning and path planning algorithms in cheerleading teaching, and provides a new technical path and theoretical basis for physical education intelligence.
