Abstract
Three-dimensional gyro-odometry has been developed to estimate the location of mobile robots on uneven terrain. To enhance accuracy, a method utilizing contact point information has been proposed, requiring contact angles for each wheel. This paper investigates how errors in these contact angles influence self-localization performance through simulations of a bicycle-type mobile robot traveling on a curved surface.
Introduction
Standard wheel odometry is highly effective on flat surfaces but becomes unreliable on uneven terrain such as ramps or bumpy fields. In 3D environments, robots typically employ gyro-odometry, combining wheel encoders with gyroscopes. An advanced version of this technique incorporates contact point information—the specific angle on a wheel's circumference where it touches the ground. This allows for a more precise calculation of the robot's instantaneous rotation axis. However, the accuracy of this method is inherently tied to the precision of the contact angle measurements.
Methodology and Simulation
The study focuses on a bicycle-type mobile robot (front-wheel steering, rear-wheel drive). The core algorithm calculates the instantaneous rotation axis ($\zeta$) by analyzing velocity vectors at the contact points of the front and rear wheels.
To test the robustness of this approach, researchers simulated the robot navigating a "Triangulated Irregular Network" (TIN) model of a smooth, curved 3D surface. Two types of sensor errors were introduced into the contact angle measurements:
- White Gaussian Noise ($\sigma$): Representing random sensor jitter.
- Systematic Bias ($\mu$): Representing constant offsets, such as those caused by miscalibration.
Results and Discussion
The simulation compared a perfect sensor baseline against scenarios involving noise and bias over a 30-second traversal.
The results revealed a critical distinction between error types:
- Resilience to Noise: When only random noise ($\sigma = 0.5^\circ$) was present, the results remained nearly identical to the error-free baseline. The integration process inherent in odometry effectively filters out high-frequency random fluctuations.
- Sensitivity to Bias: Systematic bias ($\mu \approx 0.5^\circ$ to $0.6^\circ$) caused significant divergence. The Z-axis (vertical) error was particularly sensitive, growing much faster than X or Y errors. Because the contact angle informs the algorithm about the terrain's slope, a constant angular error leads the robot to "perceive" a consistent, incorrect climb or descent, causing height estimates to drift rapidly.
Conclusion
While 3D gyro-odometry using contact point information is robust against random sensor noise, it is highly susceptible to systematic measurement bias. Future work will focus on the physical design and calibration of contact angle sensors to eliminate these offsets. Validating these findings with physical prototypes remains a priority for ensuring reliable 3D navigation in real-world uneven environments.
Summary updated