Research on interactive motion generation of character animation based on DDPG algorithm

To solve the problem that traditional methods such as Inverse Kinematics (IK) and Probabilistic Motion Graph (PMG) are not adaptive enough in dynamic environment, an end-to-end generation system based on Depth Deterministic Policy Gradient (DDPG) is proposed in this paper. The system uses Actor-Critic dual network architecture to realize continuous mapping from environmental state to joint action, and combines experience playback mechanism and soft update strategy of target network to effectively suppress training oscillation and improve convergence stability. CMU Mocap and Mixamo animation libraries were used as data sources to verify model performance in the test set and compare with IK, PMG and Proximal Policy Optimization (PPO). The results show that DDPG model Joint Trajectory Smoothness (JTS) reaches 0.96, which is 12.9% higher than IK; Motion Realism Score (MRS) is 0.94, Response Delay (RL) is as low as 41ms. The 41ms response delay encompasses the full pipeline, including environment perception, DDPG model inference, and physics simulation overhead. While this delay corresponds to approximately 24 FPS, which is lower than the 30–60 FPS (16–33ms) required for high-performance interactive applications like VR, it can still be considered near-real-time for moderately complex scenarios. On consumer hardware, further reduction in latency is feasible through model lightweighting or dedicated optimization. The cumulative reward (CR) reaches the highest value of 0.89, which highlights the smoothness and real-time advantage of DDPG model. In the dynamic interactive scene test, when the object's moving speed increased by 50%, the MRS reduction of DDPG was 7.4%, which was significantly lower than that of PPO’s 15.3%, reflecting its strong robustness to environmental disturbances such as changes in obstacle density, primarily owing to the advantage of DDPG’s multi-step forward decision-making through value function estimation. Through data preprocessing, DDPG model reasoning, interaction evaluation and simulation platform module collaboration, the system provides an efficient solution for autonomous motion generation of virtual characters in complex interactive tasks.