Machine Learning with Application to Robotics

We look at tasks that require a sequence of actions that range from single arm motions, fine manipulation and coordinated behaviors. We use knowledge acquired by observing humans performing a task or by directly guiding the robot through kinesthetic demonstrations.

We develop approaches and applications that:

  • segment a task in a sequence of meaningful actions with minimal human interference
  • extract task constraints that are key to successful execution of the task
  • make it easier to teach robots new tasks, focusing on tasks that require the skill of response to physical perturbations
  • develop an intuitive way to provide corrections and refine teaching during task execution

Video playlist