Active scope camera (ASC) is a horse-shaped disaster robot which can move on and explore narrow areas while contacting environments. The teleoperation of ASC is based on the tip camera information; therefore, an operator cannot understand the contact conditions surrounding the long body of ASC. For supporting the teleoperation of ASC, the authors have been developing the system for estimating contact position in longitudinal direction for ASC. This paper reconsider the feature values of SVM model by considering the rotation of ASC. The result showed that the model could estimate the contact conditions (contact and non-contact) with the estimation accuracy of 90% regardless of ASC rotation. In addition, we applied the learned model to unknown environments which was different from the contact condition for model learning. The estimation accuracy on unknown environments was 83%. These results suggest that the proposed estimation system is partly robust to unknown environments.