Learning Visuo-Haptic Skewering Strategies for Robot-Assisted Feeding