Talks and Poster Presentations (with Proceedings-Entry):

T. Thalmann, H.-B. Neuner:
"In-Field Calibrated Odometry for Skid-Steered Mobile Robots";
Talk: Machine Control and Guidance 2016, Vichy (invited); 2016-10-05 - 2016-10-06; in: "Machine Control and Guidance 2016", (2016), Paper ID 21, 9 pages.

English abstract:
Unmanned ground vehicles (UGVs) have gained more attention in the last few years for several applications, especially for Mobile Mapping applications. Compared to the very popular unmanned aerial vehicles (UAVs) those UGVs have some advantages, e.g. in legal restrictions or in areas with limited space for movements. Examples of applications are every kind of Indoor Mapping (e.g. for Building Information Modeling (BIM)), in mine and drilling applications, tube inspection or in forestry, when mapping trees. Skid-steered mobile robots (SSMRs) fit very well for such applications due to their robust mechanical structure and high maneuverability. Depending on the requirements of the mapping tasks it is necessary to determine the trajectory with the standard deviation of few centimeters. Wheel encoders are inherent sensors on SSMRs, and can be used in odometry and sensor fusion for localization and navigation of these mobile robots.
First, this study shows that the accuracy of the proprietary navigation solution of the used SSMR does not fulfill the above mentioned requirements. Different trajectories are analyzed for this purpose. Furthermore, an improved navigation solution which comprises two steps is proposed: first, several parametrization methods of the system state to improve filter performance, e.g. by providing better possibilities for filter tuning, are discussed. Secondly, the additional benefit of estimating slip coefficients using positioning techniques is assessed. In the last part a two-step in-field calibration routine is proposed and it is subsequently shown that external positioning support at the accuracy level of the navigation solution needs to be available only in a limited area (segment) of the trajectory.
In this work a target tracking tachymeter is used to provide ground truth trajectory to estimate the amount of slippage and to assess the performance of the existing proprietary navigation solution as well. The used parametrization methods of the system state are based on the odometry models from Wong [1978] and Aussems [1999]. Several parametrization techniques are also treated and it is shown that a different approach using one translational velocity and a differential velocity instead of the side velocities left and right respectively has a beneficial effect on filter performance and on the overall navigation solution. Conventional odometry methods are insufficient and unreliable for standalone pose determination because they do not account for the slippage between the wheels and the ground when steering the mobile robot. This is why in the 2nd improvement step of the navigation solution slippage compensating models using slip ratios left and right for tracked vehicles proposed by Endo et. al [2007] are applied for SSMRs. The model including the slip coefficients is formulated as an Extended Kalman Filter (EKF) together with tachymetry or any other positioning technique with sufficient accuracy, e.g. GNSS or RFID, depending on their availability.
A two-phase algorithm is proposed for a better usage of odometry data. The first phase is used to determine and identify slip coefficients in-situ while the robot is in the field of view of the tachymeter. Like this, local properties of the ground/soil can be examined as well as current load properties of the robot. This is the essence of the executed odometry calibration. In the second phase an autonomous drive without external support is carried out. Assuming similar underground consistency within operating area the slip coefficients can be assumed similar too, which is used to improve the standalone odometry navigation solution. Tachymetry is used in the 2nd phase to provide ground truth trajectory for evaluation, but is not included in the EKF computation.
It is shown, that the extended odometry model is able to reduce the stand-alone odometry position error by at least 50% after in-situ calibration. Furthermore contribution of calibrated odometry observations to future sensor integration (e.g. with an IMU) is improved.

Unmanned ground vehicles, Odometry, Extended Kalman Filter, Slip Coefficients

Created from the Publication Database of the Vienna University of Technology.