Let people fail! Exploring the influence of explainable virtual and robotic agents in learning-by-doing tasks