Maze training schemes are commonly used in human balance training equipment for balance training.However

the existing maze generation algorithms cannot set the difficulty attributes of the generated maze

making it difficult to realize level differentiation of the maze balance training method.Based on the depth-first search(DFS)algorithm

a random weight strategy is introduced for algorithm improvement

enabling control over the length ratio and turn ratio of the maze solution path.Using Qt and Unity3D

a multi-level balance ability maze training system with changeable attributes is designed for the balance ability test platform to meet the training needs of different trainees.Through comparative experiments

the difficulty differentiation effect of the generated maze training scenarios under different weights is verified.Human tests verify the posture synchronization effect between the maze scenarios and the real training during actual training.