Robotics·8 Months (Presented at IEEE ICCE 2025, Las Vegas)
AGNI: Adaptive Sensor Fusion Research
Hybrid Confidence-Adaptive Fusion (HCAF) Under Environmental Uncertainty — IEEE 44th ICCE
Peer-Reviewed Research Visuals PendingVerified Specification
Adaptive Sensor Fusion Research
IEEE 44th ICCE (Las Vegas) covariance convergence plots and gravel obstacle test trajectories will be supplied in the next phase.
Mathematical Formulation
K_k = P_{k|k-1} H_k^T (H_k P_{k|k-1} H_k^T + \Lambda_k R_k)^{-1}Drift Gain
~32% Less
Nav Success
~95%
Execution
50 Hz Loop
The Problem
Classical Extended Kalman Filtering assumes static measurement noise covariance matrices calibrated offline in benign conditions. When an autonomous vehicle encounters dust clouds that occlude optical lenses or loose gravel that slips drive wheels, false observations are accepted with standard confidence, corrupting vehicle state estimation.
Research Highlights
- Hybrid Confidence-Adaptive Fusion (HCAF): Formulated dynamic diagonal scaling multipliers that inflate effective innovation covariance when sensor residuals spike, smoothly diminishing weights without step discontinuities.
- IMU Vibration Coupling to DWA Planner: Closed the feedback loop between high-frequency chassis vibration spectra and local trajectory planning, dynamically constraining linear velocity over washboard terrain.
- Real-Time C++ Filter Node: Implemented the complete covariance scaling algorithm as a standalone 50 Hz ROS 2 node with deterministic computational bounds on embedded hardware.
- Peer-Reviewed IEEE Publication: Lead author on research presented and published at the IEEE 44th International Conference on Consumer Electronics (ICCE 2025, Las Vegas).
Results
Demonstrated a ~32% reduction in localization error compared to standard EKF across outdoor loose gravel and achieved a 95.5% autonomous navigation success rate under complete GPS denial.