Defence of dissertation in the field of Automation, Systems and Control Engineering, M.Sc.(Tech.) Murtaza Hazara

The title of thesis is “Incremental and Transfer Learning of Contextual Skill Models for Robots”

Motion and manipulation skill learning in humans involves several building blocks such as memorization, imitation, adaptation, optimization, and generalization. The core purpose of this thesis is to endow robots with similar efficient skill learning ca-pabilities.

The thesis proposes a framework consisting of learning from human to achieve an initial policy, optimizing the policy efficiently by guided exploration, skill model for generalizing the policy to new situations, model selection for controlling the com-plexity of skill model, and active task selection to learn a skill faster. Using these key components, learning frameworks can be constructed which provide robots with the capability to acquire motion and manipulation skills autonomously.

We studied skill learning in two contexts: in-contact and free-space motions. In brief, this thesis investigates how to: (1) learn a policy for in-contact tasks such as wood planing; (2) generalize a free-space motion policy such as basketball throwing to new situations; and (3) transfer a skill from simulation to real world.

Opponent: Professor Tamim Asfour, Karlsruhe Institute of Technology, Germany

Custos: Professor Ville Kyrki, Aalto University School of Electrical Engineering, Department of Electrical Engineering and Automation.

Thesis web page

Contact information: Murtaza Hazara, Department of Electrical Engineering and Automation, [email protected], +358503225723

  • Published:
  • Updated:
URL copied!