Learning Multimodal Contact-Rich Skills from Demonstrations Without Reward Engineering

Everyday contact-rich tasks, such as peeling, cleaning, and writing, demand\nmultimodal perception for effective and precise task execution. However, these\npresent a novel challenge to robots as they lack the ability to combine these\nmultimodal stimuli for performing contact-rich tasks. Learning-based methods\nhave attempted to model multi-modal contact-rich tasks, but they often require\nextensive training examples and task-specific reward functions which limits\ntheir practicality and scope. Hence, we propose a generalizable model-free\nlearning-from-demonstration framework for robots to learn contact-rich skills\nwithout explicit reward engineering. We present a novel multi-modal sensor data\nrepresentation which improves the learning performance for contact-rich skills.\nWe performed training and experiments using the real-life Sawyer robot for\nthree everyday contact-rich skills -- cleaning, writing, and peeling. Notably,\nthe framework achieves a success rate of 100\\% for the peeling and writing\nskill, and 80\\% for the cleaning skill. Hence, this skill learning framework\ncan be extended for learning other physical manipulation skills.\n