Kez Smithson Whitehead

Email: kez.smithson@brl.ac.uk

Supervisors: Mark Hansen & Wenhao Zhang

PhD Project Title: Investigating actor-critic based reinforcement learning methods for the control and navigation of inherently unstable robots

Research Group: Centre for Machine Vision

Biography

Prior to FARSCOPE I graduating from Loughborough University with MEng Systems Engineering (First Class) and worked in industry for a year. I became fascinated by how recent model free algorithms; such as Proximal Policy Optimisation, could allow simulated humanoids to learn to walk from random behaviour resulting in gaits that ‘almost’ revival our own. My research aims to builds on this by exploring some of the issues that arise with apply reinforcement learning to real world robotics including:

• Sample efficiency

• Automatic post terminal state resets

• Damage avoidance

• Rapid adaptation to environment changes

• Failure Traceability

I enjoy designing, building and programming torque compliant robots as well as adding user friendly interfaces for my c and c++ code.

I am also one of the founders of the Bristol Reinforcement Learning reading group, which has been going strong for several years and we are always happy to welcome new members, please contact myself or Mihai for details!