In 2025, my group and I are starting an ambitious new project in robot learning for multi-fingered, dexterous manipulation. We will be studying the intersection of reinforcement learning, evolutionary algorithms, and sim-to-real transfer, and I am now looking for outstanding candidates for the following fully-funded positions:
 

• 2 x 4-year PhD positions, for any nationality, to begin in October 2025. These positions are ideal if you would like to study in depth the intersection of reinforcement learning, evolutionary algorithms, and sim-to-real transfer, within the context multi-fingered dexterous manipulation.

• 1 x 4-year Post-Doc position, for any nationality, to begin between April and June 2025. This position is ideal if your PhD included significant experience working on robot learning algorithms with evaluation on real-world robots and real-world tasks, and you would like to help me lead a new multi-disciplinary team across both robotics and machine learning.

• 1 x 4-year Research Assistant position, for any nationality, to begin between April and June 2025. This position is ideal if you are a recent graduate from an undergraduate or master's degree, and you are looking to gain some hands-on experience in an academic research lab, before applying for a PhD or other research positions.

​

Specifically, in this project we will be studying co-design of robot hardware and control policies, which aims to simultaneously optimise the robot hardware (e.g. the number of fingers, the sizes, shapes, and positions of the fingers, and the positions of the tactile sensors), and the control policy for that hardware (e.g. a policy which takes in proprioceptive and tactile observations, and outputs target joint positions). We will be investigating evolutionary algorithms for optimising the hardware, using population-based search methods to explore the full range of possible hardware configurations. For policy learning, we will be studying reinforcement learning to incrementally optimise the policy each time a hardware optimisation step is taken. Together, both the hardware and policy will be optimised jointly by taking small steps in each, whilst fixing the other. All of this will be done within a huge GPU-based simulation with thousands of robots learning synchronously in parallel. The best solutions will then be rapidly 3D-printed, and policies will be fine-tuned with reinforcement learning in a robot "arm farm" we are building at Imperial College.

Through this project, we are aiming to automatically generate creative new robot hands and highly dexterous control policies, for human-level or even super-human performance across a range of everyday tasks. This is part of a large collaboration with Nathan Lepora at the University of Bristol, together with several other organisations both in the UK and abroad.

 

I will be accepting official applications soon. To help me manage the number of applications, please let me know of your intention to apply by completing the relevant form below, by 14th March 2025:
 

• PhD positions: Complete Form Here (Note: if you are also applying for regular PhD positions in my lab, please also complete the other form linked to at www.robot-learning.uk/phd-applications)

• Post-Doc position: Complete Form Here

• Research Assistant Position: Complete Form Here

​

After receiving your form submission, I can provide some brief feedback as to whether your profile fits the position. For promising candidates, I will then encourage you to submit an official application, after which I will invite the top candidates for interviews.

 

For informal enquiries about any of these positions, please contact me at e.johns@imperial.ac.uk.

I look forward to hearing from you!

Edward Johns,
Director of the Robot Learning Lab,
Imperial College London.