Fault-Tolerant Six-DoF Pose Estimation for Tendon-Driven Continuum Mechanisms

We propose a fault-tolerant estimation technique for the six-DoF pose of tendon-driven continuum mechanisms using machine learning. In contrast to previous techniques, no deformation model is required, and prediction rather performed with polynomial regression. As only few datapoints are required regression, several estimators trained structured occlusions available sensor information, clustered into ensembles based on sensors. By computing variance one ensemble, uncertainty in monitored and, if above threshold, loss detected handled. Experiments humanoid neck DLR robot DAVID, demonstrate that accuracy predicted significantly improved, reliable can still be 3 out 8