Abstract
To estimate the position and heading angle of mobile robot precisely, an measurement variable estimation model was proposed to adapt any angle. Fusing the predictive value of odometry and measurement data of multiple sensors by the Extended Kalman Filtering (EKF) for reducing the accumulative error by using only traditional odometry. The proposed models is verified by Matlab simulation and experimental results.
Introduction
For an autonomous robot to navigate effectively, precise self-localization is paramount. Most mobile robots rely on wheel odometry—calculating position based on wheel encoder data—as a primary localization source. However, odometry suffers from inherent "drift" or accumulative error caused by wheel slippage, mechanical tolerances, and uneven surfaces. Over time, these small errors compound, leading to significant discrepancies between the robot's estimated and actual positions.
This research addresses the drift problem by implementing a sensor fusion approach. By combining high-frequency odometry data with absolute distance measurements from a Time-of-Flight (ToF) laser sensor via an Extended Kalman Filter (EKF), the system provides a stable and accurate estimate of the robot's position $(x, y)$ and heading angle $(\theta)$.
Methodology: The EKF Framework
The core of the localization system is the Extended Kalman Filter, which serves as a recursive estimator for the robot's state. The process is divided into two main phases: prediction and correction.
Kinematics Modeling (Prediction)
The researchers first established a mathematical model of the tracked robot's movement. By monitoring the angular velocities of the left and right tracks, the system predicts the robot's next state. While responsive, this prediction step is where the accumulative error typically originates.
Measurement Modeling (Correction)
To correct the predicted state, the robot utilizes a VL53L0X ToF sensor mounted on a servo motor. This sensor scans the environment across a 180° arc. A key innovation in this paper is a universal measurement model that adapts to obstacles at any angle. This model calculates the expected distance to known environmental features (like walls) and compares it to the real-time sensor readings.
The EKF algorithm linearizes these non-linear movement and measurement equations using Taylor series expansion. It then calculates the "Kalman Gain"—a weighting factor that determines whether to trust the internal odometry or the external laser measurements more at any given moment—to produce the final optimal estimation.
Experimental Setup
The proposed method was validated through both Matlab simulations and physical experiments. The hardware platform consisted of a custom tracked mobile robot equipped with:
- Wheel encoders for odometry.
- A VL53L0X ToF laser-range-finder (LRF) mounted on a servo.
- An onboard Microcontroller Unit (MCU) for real-time processing.
- A Bluetooth module for data logging and monitoring.
Testing was conducted in a controlled laboratory environment with rectangular obstacles, allowing the researchers to compare the EKF-estimated path against the ground truth.
Key Results and Insights
The experimental results confirmed that the EKF-based approach significantly outperforms standalone odometry.
- Precision: The system achieved a localization error of less than 1.01 cm across 300 test points.
- Drift Mitigation: While the odometry-only path diverged over time, the EKF estimation remained stable, effectively "anchoring" the robot to its true coordinates using the laser measurements.
- Flexibility: The universal angle adaptation of the measurement model allowed the robot to handle obstacles at various orientations, moving beyond the 90° limitations of simpler models.
Conclusion and Future Work
This study demonstrates that high-precision localization can be achieved using affordable sensors and robust filtering algorithms. By fusing "drifty" odometry with absolute ToF measurements, the researchers successfully mitigated accumulative error in a tracked robot.
However, certain limitations remain. The current model assumes a flat surface and static obstacles. Future research will focus on adapting the system for dynamic environments with moving obstacles and investigating the effects of significant wheel slippage on rougher, non-ideal terrains.
Summary updated