The Joint Adaptive Kalman Filter (JAKF) for Vehicle Motion State Estimation

Area of Science:

• Robotics
• Sensor Fusion
• Autonomous Systems

Background:

• Accurate estimation of vehicle motion is critical for advanced driver-assistance systems (ADAS) and autonomous driving.
• Traditional Kalman filters struggle with varying sensor noise and potential sensor failures, impacting overall system performance.
• Existing adaptive Kalman filters offer improvements but may not fully address multi-sensor fusion challenges.

Purpose of the Study:

• To propose a novel multi-sensory Joint Adaptive Kalman Filter (JAKF) for robust vehicle motion state estimation.
• To enhance data fusion by adaptively adjusting noise variance-covariance matrices (R and Q) using Lidar and Radar inputs.
• To improve the accuracy and fault tolerance of motion estimation compared to conventional methods.

Main Methods:

• Extended innovation-based adaptive estimation (IAE) to develop the JAKF.
• Utilized Lidar and Radar data as inputs for local filters to adaptively adjust measurement noise (R) and system noise (Q) V-C matrices.
• Implemented a global filter for optimal data fusion using an information allocation factor (β) derived from R, with feedback to local filters.

Main Results:

• JAKF demonstrated superior adaptive ability and fault tolerance in extensive simulations and experiments.
• The filter effectively bridged accuracy differences between Lidar and Radar sensors, improving overall filtering effectiveness.
• JAKF maintained a stable convergence rate even when individual sensors failed, outperforming conventional Kalman filter (CKF) and innovation-based adaptive Kalman filter (IAKF) in displacement, velocity, and acceleration accuracy.

Conclusions:

• The proposed JAKF offers a significant advancement in multi-sensor fusion for vehicle motion estimation.
• JAKF provides enhanced accuracy, adaptive noise handling, and improved robustness against sensor failures.
• This approach is highly promising for real-world ADAS and autonomous driving applications requiring reliable state estimation.