Learning contact-rich manipulation policies typically requires large amounts of human demonstration data. Robot-free collection with handheld interfaces has emerged as a scalable way to gather such demonstrations, letting humans perform tasks naturally without being tethered to a robot.

However, the abstract notes that existing handheld interfaces often provide limited intuitive force feedback. This is a significant drawback for contact-rich tasks, where the feel of forces and contacts is essential for successful execution.

To address this, the paper introduces FOMI, a force-feedback manipulation interface. The work suggests that by providing more intuitive force feedback, FOMI could improve the quality of demonstrations and, in turn, the performance of learned policies in contact-rich manipulation.