As quadruped robots are increasingly applied in complex environments, how to efficiently control their motion has become a critical research issue. To enhance the performance of the robot control system, this study optimizes the Deep Deterministic Policy Gradient (DDPG) algorithm by integrating a Convolutional Neural Network (CNN) to improve the adaptability and motion control efficiency of quadruped robots in dynamic environments. Experimental results demonstrate that CNN-DDPG excels across multiple evaluation metrics. In terms of convergence speed, CNN-DDPG achieves an average reward of 9.5 within the first 100 iterations, significantly higher than the 5.2 for DDPG, 6.3 for Proximal Policy Optimization (PPO), 4.9 for Asynchronous Advantage Actor-Critic (A3C), and 7.0 for Soft Actor-Critic (SAC). As the number of iterations increases, the performance of CNN-DDPG continues to improve, notably outperforming other algorithms. Additionally, CNN-DDPG exhibits lower loss values and shorter path lengths in loss function optimization and path planning, especially in complex obstacle environments, effectively reducing redundant paths and enhancing task execution efficiency. By introducing CNN to optimize the DDPG algorithm, CNN-DDPG significantly improves the convergence speed and optimization effect of the robot control system in complex tasks, demonstrating high practical value and application prospects.