Master thesis on “Hybrid learning and control for human-robot interaction: an imitation learning perspective”

Supervisor: Prof. Ville Kyrki (ville.kyrki@aalto.fi).
Advisors: Dr. Fares J. Abu-Dakka (fares.abu-dakka@aalto.fi), Dr. Yanlong Huang (Y.L.Huang@leeds.ac.uk).

Keywords: imitation learning, kernelized movement primitives, robot learning and manipulation, learning from demonstration, hybrid control.

Project description

Imitation learning provides a straightforward way to transfer human skills to robots. In this context, plenty of works have been carried out to learn various types of skills, e.g., in the form of robot Cartesian position/orientation, joint position/velocity, stiffness and force profiles. This project will move further to exploit more comprehensive information from humans, aiming to learn as many skill patterns from humans as possible according to the tasks at hand. Specifically, this project will leverage the state-of-the art approach Kernelized Movement Primitives to address both the learning and control problems associated with time input and high-dimensional inputs. This project is expected to be implemented in a bunch of robot platforms and provide a user-friendly interface to users in a wide range of applications, including single arm manipulation, dual-arm task and human-robot collaboration.

Hint: For starting up with coding, the student is advised to refer to our code repository in github.

Deliverables

  • Review relevant imitation learning literature;
  • Review relevant hybrid learning and control algorithms;
  • Implementation of relevant algorithms;
  • Design evaluation setups and implement the method on real robots.

Practical information

Prerequisites: Background in machine learning, applied mathematics, ROS, Linux.
Suggested tools: C++, Matlab, or Python; and for simulation: MuJoCo or VRep.
Platform: Franka Panda robotic arm, Kuka LWR.
Start: Available immediately

References

[1] Yanlong Huang, Leonel Rozo, João Silvério and Darwin G. Caldwell, Kernelized Movement Primitives, International Journal of Robotics Research, 2019. [video]
[2] Yanlong Huang, Fares J. Abu-Dakka and João Silvério, Darwin G. Caldwell, Towards Orientation Learning and Adaptation in Cartesian Space, IEEE Transactions on Robotics, 2020. [video-1] [video-2] [page]
[3] Abu-Dakka, Fares J., Bojan Nemec, Jimmy A. Jørgensen, Thiusius R. Savarimuthu, Norbert Krüger, and Aleš Ude. “Adaptation of manipulation skills in physical contact with the environment to reference force profiles.” Autonomous Robots, 2015. [video-1] and [video-2]