Opportunity: Task and Motion Planning for Manufacturing Assembly

Modern manufacturing environments increasingly use robots equipped with advanced cameras and force/torque sensors to adapt to changes in the environment. While these sensors generate massive amounts of raw, “low-level” data when observing robotic tasks, this unorganized data is not separated into reusable and human-interpretable parts. To help robots apply actions across different manufacturing tasks, we need to translate raw data into higher-level descriptions known as abstractions. For example, a sudden spike in force data isn’t directly actionable on its own. However, if we can map that spike to an abstraction such as “obstruction-encountered”, the robot can select an action like “move-around-obstruction” that worked in previous, similar states. Selecting the appropriate abstractions for robot planning is a complex challenge. In this opportunity, you will investigate which abstractions are useful for task and motion planning in multi-modal manufacturing tasks, such as drone assembly. You will explore, compare, and implement AI methods to generate these abstractions. Thus, this is a great chance to get experience at the intersection of robotics and artificial intelligence.