Yoojin Oh

Yoojin Oh

yoojin.oh@ipvs.uni-stuttgart.de

Teaching

Available Thesis Topics

Shaping arbitration using reinforcement learning

In shared control, the policies of the user and the autonomous robot agent are blended using an arbitration function. We aim to learn an arbitration function for shared control that adapts to the user while the user interacts with the system using reinforcement learning. The arbitration function is learned and customized online to the user depending on the user’s preference to robot assistance. We focus on a teleoperation pick-and-place setting where the user controls the Baxter robot arm using a joystick. Good knowledge in Reinforcement learning, machine learning, and python is required.

Shared control policy using predicted intent priors

We aim to predict the grasp location and orientation for the object during robot pick-and-place teleoperation. The predicted position/orientation will be used as a prior for the robot policy to control the robot using Shared Control. Machine learning methods including Mask-RCNN and supervised learning will be used. Good knowledge of machine learning, optimization, and python is required. Students who took “Practical Course Robotics” are preferred.

Lectures (TA)

Previous

  • (Fall 2015, Spring 2017) Robotics
  • (Spring 2016, Fall 2016) Dynamics