Teaching Robots Interactively (TERI)
Programming and re-programming robots is extremely time-consuming and expensive. This presents a major bottleneck in new industrial, agricultural, care, and household robot applications. The goal is to realise a scientific breakthrough that will enable robots to learn how to perform manipulation tasks from a few human demonstrations, based on novel interactive machine learning techniques.
Current robot learning approaches focus either on imitation learning (mimicking the teacher’s movement) or on reinforcement learning (self-improvement by trial and error). Learning even moderately complex tasks in this way still requires either infeasibly many iterations or task-specific prior knowledge that needs to be programmed in the robot. TERI will seek to make robot learning faster, and more effective and efficient by incorporating intermittent robot-teacher interaction. This has so far been largely ignored in robot learning, although it is a prominent feature in human learning. The project will deliver a completely new and better approach: robot learning will no longer rely on initial demonstrations only, but will rather use additional user feedback effectively to continuously optimise task performance. It will enable the user to directly perceive and correct undesirable behaviour and to guide the robot quickly toward the target behaviour. Previous research has included ground-breaking contributions to existing learning paradigms, and TERI is especially well ideally prepared to tackle its three-fold challenge, developing theoretically sound techniques which are at the same time intuitive for the user and efficient for real-world applications.
The novel framework will be validated via generic real-world robotic force-interaction tasks related to handling and (dis)assembly. The potential of the newly developed teaching framework will be demonstrated with challenging bi-manual tasks and a final study evaluating how well novice human operators can teach novel tasks to a robot. This project is an ERC Starting Grant of Jens Kober, 2019-2023.