LiSA-Nav: Onboard LiDAR-Only Socially Aware Predictive Navigation for Quadruped Robots

Socially aware navigation is a critical capability for robots operating in human-populated environments, where safety must be ensured while respecting implicit social norms. This paper presents LiSA-Nav, a fully onboard, LiDAR-only socially aware predictive navigation framework for quadruped robots. The proposed system detects and tracks pedestrians directly from LiDAR point clouds, predicts their short-horizon motion, and explicitly incorporates these predictions into a socially aware local planner to enable anticipatory and socially compliant navigation behaviors. The robot can proactively yield in confined spaces, maintain appropriate interpersonal distances, and avoid abrupt or intrusive maneuvers. All perception, prediction, and planning modules run entirely onboard a single NVIDIA Jetson Xavier NX, without reliance on cameras, external computation, or learned social policies. Extensive real-world experiments in both indoor and outdoor environments, including doorway interactions, corridor encounters, and outdoor pedestrian scenarios, demonstrate that incorporating human trajectory prediction increases the minimum human--robot distance by up to 0.25m and reduces socially intrusive interactions within interpersonal distances by up to 26% compared with geometric-only LiDAR navigation, while maintaining comparable time in head-on encounters and reducing time by up to 13% in crossing scenarios. These results validate LiSA-Nav as an effective and practical solution for polite, predictable, and safe quadruped robot navigation in shared human environments.