Vision- and tactile-based continuous multimodal intention and attention recognition for safer physical human-robot interaction