Researchers have created a benchmark called The Imitator Game to test how well robots can imitate human actions. They've also developed a large dataset of paired human-robot interactions and an evaluation platform for testing robot imitation abilities. The results show that current state-of-the-art models struggle to understand the intent behind human demonstrations, especially when faced with new tasks or environments. The study suggests that fine-tuning these models with just a few examples can significantly improve their performance. --- Why it matters: This matters because it highlights the limitations of current robot imitation techniques and points to the need for more sophisticated approaches that can understand human intent. Engineers working on human-robot interaction will be interested in this research as it provides insights into how to improve robot learning and imitation abilities. Source: https://arxiv.org/abs/2608.22301