A Safe Data-Driven Optimization Approach for Robot Navigation in Dynamic Environments

This paper presents a novel, data-driven motion planning strategy for autonomous mobile robots navigating in dynamic environments with human interactivity. The proposed approach utilizes a receding-horizon optimization framework that integrates predictive models of the robot motion with unknown-form safety constraints encapsulating human movement uncertainties, and complex dynamics of human-human and human-robot interactions. The functional form of constraints is unknown instead, we obtain only measurements and gradients of the constraint i.e. 1st order online optimization. Specifically, data-driven log barrier functions enforce safety constraints by penalizing closeness to constraint boundaries. The proposed strategy enables reliable, efficient, and safe robot navigation in high-density environments, making it particularly suitable for applications in pedestrian and other interactive spaces. Finally, realistic simulation studies validate the effectiveness of the proposed framework in balancing safety with operational efficiency.