Shared Trajectory-Based Multi-Policy Decision-Making for Socially Compliant Robot Navigation in Dense Crowds

Robot navigation must adhere to human social norms in dense pedestrian crowds to better integrate into human daily life. This paper proposes a shared trajectory-based multi-policy decision-making framework that addresses limitations in modeling implicit social norms and handling unstable policy transitions in existing methods. A compact asymmetric interaction-space model is used to describe pedestrian personal and group spaces, and side-preference norms are explicitly incorporated into decision-making. Multiple policies, including solo motion, following, and stopping, are generated from a shared set of sampled paths via different temporal allocations, while an additional rotation-based recovery policy is introduced for frozen or temporarily blocked states. Candidate trajectories are evaluated and selected in terms of comfort, efficiency, social-norm compliance, consistency, and smoothness. Simulation and real-world results in corridor, crossing, and indoor scenarios show that the proposed method improves stability, safety margins, and efficiency over representative baselines. Note to Practitioners - This work is motivated by the practical need for mobile robots to comply with human social norms when navigating in crowded public environments. In real deployments, robots must deal with dynamically changing pedestrian behaviors and implicit conventions-such as preferred walking sides-which are difficult to explicitly encode in existing navigation systems. Existing methods often focus mainly on collision avoidance and fail to fully support socially-compliant navigation in dense crowds. This paper proposes a practical social navigation solution suitable for service robots, delivery robots, and other mobile platforms operating closely with humans. The approach quantifies common social norms and employs a compact asymmetric Gaussian mixture model to represent interaction spaces in a unified manner. In addition, a shared trajectory-based decision-making strategy is employed, allowing different pedestrian motion patterns-such as stopping, following, or merging into flows-to be mimicked. Simulation and real-world experiments demonstrate that the proposed method improves both comfort and efficiency during navigation. Nonetheless, its performance depends on environment-specific weighting parameters.