Abstract
Intelligent vehicles in autonomous driving and obstacle avoidance require precise relative state estimation. For vehicle-borne sensor networks with time-varying transmission delays, this paper proposes an anti-delay distributed Kalman filter (DKF) with finite-time convergence. By integrating a consensus strategy with buffer technology, the algorithm achieves optimal fusion in the sense of linear minimum variance. The approach accommodates practical constraints like communication weight coefficients and directed topologies. Theoretical proofs for error covariance upper bounds and maximum allowable delays are provided. Simulations and real-world experiments verify that the proposed FtDKF algorithm offers superior accuracy and robustness compared to existing methods.
Introduction
In autonomous driving, precision is paramount. Vehicles rely on sensor networks to track position, velocity, and acceleration. While centralized filtering is accurate, it introduces computational bottlenecks and single points of failure. Distributed Kalman Filters (DKF) allow nodes to perform local estimations and reach consensus via "gossip" protocols. However, real-world networks face two major hurdles: transmission delays and slow convergence. High-speed vehicles require reliable state estimates within a finite timeframe, a luxury not afforded by traditional asymptotic consensus algorithms.
Methodology: The Anti-delay FtDKF
The proposed Anti-delay Finite-time Distributed Kalman Filter (FtDKF) architecture rests on three pillars:
Buffer Technology
To neutralize the impact of out-of-order or delayed packets, each sensor node maintains a data buffer. This allows the system to store and re-order incoming neighbor data based on timestamps before fusion.
Finite-Time Consensus and Optimal Weighting
Unlike standard filters, this algorithm reaches the optimal fusion state in a finite number of iterations related to the network's diameter. It utilizes a matrix weighting approach based on linear minimum variance, ensuring that more "trustworthy" sensor data (with lower covariance) carries more weight.
Directed Topology Support
The algorithm is proven to work on strongly connected digraphs (unidirectional communication), which better reflects real-world V2X scenarios where signal paths may be blocked.
Experimental Validation
The researchers employed two validation methods:
- Numerical Simulation: A 12-sensor network tracked a target's state under varying transmission delays ($d_t = 1$ to $4$).
- Hardware Case-Study: Three 24G millimeter-wave radar sensors (Nanoradar SP25) tracked a mobile car at $T=0.01s$ intervals.
Key Results and Insights
The FtDKF consistently outperformed recent benchmarks, such as the 2020 work by Yang et al.
- Superior Accuracy: As shown in the Mean Squared Error (MSE) plots, the proposed algorithm maintains significantly lower error across position ($s$), velocity ($\dot{s}$), and acceleration ($\ddot{s}$).
- Stability Under Delay: While competing algorithms experienced error spikes as delays increased, the FtDKF remained stable.
- Maximum Delay Boundary: A key contribution is the derivation of the maximum allowable delay a network can tolerate before accuracy degrades beyond acceptable thresholds.
Conclusion and Future Work
The FtDKF provides "global-level" accuracy without the centralized computational burden, making it highly suitable for practical vehicle applications. Future research will focus on extending the framework to non-linear system dynamics (e.g., EKF/UKF) to handle emergency maneuvers and investigating scalability for massive vehicle platoons.
Summary updated