top of page
List of projects for MSc 2018




Robotic Grasp Selection with Deep RL

Description: Learn how to evaluate grasps on novel objects by using DRL. The learn reward function should be parametrised by the task, so that the robot can cope with task constraints.

Requirements: Strong math background & strong coding skills (preferably C++)

Framework: In-house C++ library (Golem), Kernel Density Estimation, Deep RL

N. Students: max 2


Feature-based Transferable Robotic Pushing Predictors

Description: We aim to develop a predictor that can learn how objects behave under push operation. The goal is to learn a transferable set of features that will enable the robot to make predictions on different shaped objects. Use of such predictors in frameworks like MDPs to find optimal policies.

Requirements: Strong math background & strong coding skills (preferably Python and/or C++)

Framework: ROS, Kernel Density Estimation, RL, Stochastic Optimal Control

N. Students: max 4


Robotic Grasping of Moving Objects

Description: We aim to improve the grasping abilities of a robot manipulator by enabling it to make predictions on how to grasp moving objects (i.e. roller conveyor)

Requirements: Strong math background & strong coding skills (preferably Python and/or C++)

Framework: In-house C++ library (Golem), Kernel Density Estimation, Linear Algebra

N. Students: max 2




Efficient Action-Selection Strategies in Sparse Problems

Description: How can we efficiently learn to explore in problems where the reward states are sparse, i.e. Atari Montezuma's revenge? We aim to investigate different directions to improve the applicability of RL techniques to real-world problems.   

Requirements: Strong math background & basic-level coding skills (preferably Python)

Framework: DeepMind StartCraft II API, RL, Deep RL, Recurrent Q-Networks

N. Students: max 4




Towards the Understanding of Human Learning (with Prof. Aaron Sloman)

Description: How ancient Greek mathematicians have been able to understand and reason about topological and geometrical concepts before Euclid's propositions?  

Requirements: Philosophy and/or neuroscience background

Framework: Modelling of human behaviour, basic math background

N. Students: max 2




Development of a Graphical Tool for Teaching AI

Description: Development of a platform to visualise AI algorithms for the Hero-Seeking-Treasure problem. The students will improve their knowledge of the AI algorithms (branch & search, MCTS, Minimax, Belief Nets, Neural Nets, and Value & Policy iteration) seen in the lectures by applying them to the same toy problem. It is required to build the GUI as well. The final product should be an interactive platform (API) in which next years' students can write their own version of the algorithms. Possible extension to learn self-checking of the code and marking.

Requirements: Basic/beginner coding skills

Framework: Python

N. Students: max 5

bottom of page