Context-Based Echo State Networks with Prediction Confidence for Human-Robot Shared Control