Uncertainty-Aware Multi-Sensor Fusion and Adaptive Receding-Horizon Path Planning for Autonomous Navigation in Dynamic Outdoor Environments

Background: Outdoor autonomous navigation faces simultaneous challenges of unreliable Global Navigation Satellite System (GNSS) reception, perceptual aliasing, and rapidly changing obstacle fields. While sensor fusion and dynamic path planning have each matured, their tight coupling through uncertainty-aware objectives remains underexplored in dynamic outdoor scenes. Methods: We propose an integrated architecture that combines a robust, adaptive multi-sensor fusion backbone with a predictive, multi-objective receding-horizon planner. The fusion module fuses inertial, wheel, camera, LiDAR, and GNSS observations using an error-state extended Kalman filter with innovation-based adaptive noise estimation and zero-velocity updates, emphasizing modularity and robustness (Lynen et al., 2013; Jwo & Weng, 2008; Ma et al., 2018). The local planner optimizes a kinodynamically feasible trajectory under a risk-aware cost that leverages fused-state covariance, time-to-collision predictions, and energy terms inspired by multi-objective navigation (Mandow et al., 1998; Ge et al., 2007; Ferrer & Sanfeliu, 2018). A topological, Voronoi-skeleton global guide provides wayfinding under continual re-planning (Lee & Song, 2004; Brenner & Nebel, 2009). Results: In twelve outdoor runs (8.1 km total) across pedestrian-dense campus paths, the proposed system achieved 96.7% task completion without intervention and 0 collisions. RMS localization error decreased by 34% compared to the best non-adaptive baseline. Minimum clearance and time-to-collision improved while maintaining higher average speed. Ablation shows adaptive measurement noise estimation reduced drift and near-miss rates in GNSS-challenged segments. Conclusions: Uncertainty-aware coupling of fusion and planning materially enhances safety and efficiency in dynamic outdoor navigation. The approach complements prior advances in sensor fusion, kinodynamic planning, and continual re-planning, and invites extensions to multi-robot cooperation and aerial platforms (Indelman, 2017; Shakhatreh et al., 2019).