State Estimation for Snake Robots

David Rollinson, Austin Buchan, and Howie Choset
IEEE International Conference on Intelligent Robots and Systems (IROS), October, 2011.


Abstract
We present a comparison of methods to estimate the shape and orientation of a locomoting snake robot by fusing the robot’s redundant internal proprioceptive sensors using an Extended Kalman Filter (EKF). All of the estimators used in this work represent the shape of the snake with gait parameters to reduce the complexity of the robot configuration space. The compared approaches for representing shape and pose of the snake robot differ primarily in the use of a body frame fixed to the pose of a single module versus one that is aligned with the virtual chassis. Additionally, we evaluate a state representation that explicitly tracks joint angles for improved estimates. For one particular gait, rolling, we present experimental data where motion capture data of the snake robot is used as ground truth to compare the accuracy of the state estimates from these techniques. We show that using the virtual chassis body frame, rather than a fixed body frame, results in improved accuracy of the snake robot’s estimated pitch and roll. We also show that, in general, representing the robot’s shape with gait parameters is sufficient to accurately estimate shape and pose, though it can be improved upon in specific cases by explicitly modeling joint angles. I.

Notes

Text Reference
David Rollinson, Austin Buchan, and Howie Choset, "State Estimation for Snake Robots," IEEE International Conference on Intelligent Robots and Systems (IROS), October, 2011.

BibTeX Reference
@inproceedings{Rollinson_2011_7381,
   author = "David Rollinson and Austin Buchan and Howie Choset",
   title = "State Estimation for Snake Robots",
   booktitle = "IEEE International Conference on Intelligent Robots and Systems (IROS)",
   month = "October",
   year = "2011",
}