ON-LINE ESTIMATION OF SKID-STEER INSTANTANEOUS CENTERS OF ROTATION IN GPS-DENIED ENVIRONMENTS

摘要

The vast majority of explosive ordinance disposal and hazardous area inspection ground robots utilize skid-steer locomotion technology due to the design's robustness, simplicity, and maneuverability advantages. Autonomous algorithms are difficult to adapt to these platforms because dynamic models predicting skid-steer movement are complex and rely heavily on prior knowledge of surface and vehicle parameters. This paper will discuss the use of skid-steer instantaneous center of rotation (ICR) kinematics for predicting robot motion with limited prior knowledge of vehicle design or surface condition. Prior work has shown that ICR kinematics can be estimated using an extended Kalman filter (EKF) when measurements of input track/wheel speed, position, and heading are available. The resulting ICR estimates were shown to produce accurate predictions of vehicle movement; much more accurate than a naive two-wheel robot kinematic approach using identical inputs. The results for two implementations of an ICR EKF using measurements obtained from a LIDAR based simultaneous localization and mapping (SLAM) algorithm in one case and measurements from a stereo visual odometry algorithm in the second will be presented. Results show that the SLAM algorithm, implemented on a wheeled skid-steer platform, and the visual odometry algorithm, implemented on a tracked platform, produce ICR estimates which accurately model the motion of the vehicle. Such ICR estimates can then be used for model-predictive control, path planning, and other improvements to skid-steer autonomy.