Bachelor projects AI (RUG)

Bachelor Project with Dr. Marco Wiering

• Reinforcement Learning for POMDPs.

  
  
  Reinforcement learning (RL) enables an agent to learn to act in an unknown environment by interacting with it. Here we want to use RL for solving partially observable Markov decision processes, which are in general considered to be the hardest single agent control problems. Dr. Marco Wiering has invented some new RL algorithms for this purpose, and the goal of the bachelor student is to compare that algorithm to the Q-MDP algorithm. Q-MDP is based on first using a dynamic programming algorithm and has been shown to be successful for solving particular POMDPs. Therefore, the student will work with existing software written in C++, implement the Q-MDP algorithm, and compare that algorithm to the ones Dr. Marco Wiering has already implemented on some POMDP problems such as navigation under high sensor uncertainty. More information can be obtained by sending an email to Marco Wiering