Vehicle-to-vehicle cooperative localization using robust HOC M-estimation with VIO and distance constraints

Vehicle-to-vehicle (V2V) cooperative localization mitigates the cumulative drift of standalone localization by exploiting position and ranging information from nearby vehicles. Compared with existing approaches that are based solely on inter-vehicle distance constraints, combining visual-inertial odometry (VIO) with ranging measurements improves accuracy while reducing the dependence on the number and density of cooperative vehicular anchors. Such methods are particularly promising for applications in intelligent transportation systems (ITS) and GNSS-degraded environments. Building on this, this paper proposes a robust V2V cooperative localization method that combines multi-epoch cooperative observations with visual-inertial odometry (VIO) pose priors in a batch weighted least-squares framework. By exploiting multi-epoch inter-vehicle ranging constraints and onboard VIO pose estimates, the proposed approach improves localization accuracy while reducing the reliance on dense cooperative participation, and remains effective even in two-vehicle scenarios. A robust M-estimator with a hybrid ordinary-cauchy (HOC) kernel is employed to adaptively weight the VIO results and cooperative measurements by jointly considering theoretical observation reliability and actual measurement quality, thereby reducing the influence of low-quality observations on cooperative localization. A systematic theoretical analysis is then conducted and validated through simulations, including error distribution modeling, Cramér–Rao lower bound (CRLB) derivation, and an evaluation of the accuracy gains from VIO fusion. Finally, real-world experiment results demonstrate that the proposed method achieves higher positioning accuracy and robustness than the standalone VIO and representative cooperative localization baseline methods.