Abstract
Reliable navigation for underground mining vehicles remains a critical challenge due to the rapid error accumulation of low-cost micro-electro-mechanical systems sensors in global navigation satellite systems-denied environments. To reconcile hardware limitations with high-precision requirements, this article proposes the adaptive federated Kalman filter (A_FKF), a fusion framework wherein a pyramid-shaped volumetric topology combined with singular value decomposition kinematics is constructed. By rigorously modeling the lever-arm effect, this approach explicitly decouples rotational dynamics from translational acceleration to resolve the observability deficiencies of single-point sensing. Integrated within this framework, a decentralized federated Kalman filter architecture featuring dynamic information sharing is developed, which reallocates fusion weights based on real-time subsystem confidence to mitigate the impact of local sensor inaccuracies. Validation under a 300 s dynamic trajectory confirms the algorithm effectively suppresses cumulative drift, achieving a 1.480 m position root mean square error (RMSE) and 3.214 m 95% spherical error probable. Furthermore, a regular pyramid sensor topology mitigates coplanar geometric dilution of precision; compared to planar baselines, this volumetric configuration restricts attitude errors below 0.5° and further reduces position RMSE to 1.131 m. These results substantiate the method’s efficacy for robust, low-cost autonomous navigation in complex mining environments.
Funding Information
  • Natural Science Foundation of Anhui Province (2308085ME150)
  • Anhui Province Outstanding young talent project (YQYB2024048)
  • National Natural Science Foundation of China (52574188)
  • Fundamental Research Project of Chuzhou (JXWD202201)
  • Anhui Province Postdoctoral Fund Project (2025B1062)

This publication has 37 references indexed in Scilit: