<p>Safe and socially compliant navigation in dense human crowds presents a critical challenge for mobile robots, requiring consideration for not only physical safety but also human psychological comfort. However, current approaches often overlook the social impact of a presence of the robot on nearby people and fail to adequately model the temporal nature of navigation tasks. This limits their robustness in dynamic and uncertain environments. To address these limitations, this paper proposes a novel social navigation algorithm: Structured Neural Circuit Policies with Proximal Policy Optimization (SNCP-PPO) algorithm. First, we design a biologically-inspired Actor-Critic network that uses ordinary differential equations (ODEs) to model the continuous-time evolution of states. This captures temporal causality in interactions and significantly enhances the decision-making robustness of the model. Second, we establish a human social space model and introduce a social pressure index to quantify the discomfort caused by the robot. By integrating this index into the reinforcement learning reward function, the robot learns to minimize its social intrusion while maintaining navigation efficiency. Experimental results in both simulated and real-world settings demonstrate that SNCP-PPO algorithm achieves a 93% to 95% success rate in dynamic environments with 10 to 20 people. It also outperforms baseline algorithms and generates more socially compliant trajectories.</p>