Programmer. Hiker. Cook. Always looking for interesting problems to solve.
Final year project completed
14 May 2013 |
It’s been a long road, but the project is finally completed: I submitted report and survived my viva
To sum up:
- Project was part of a study on cognitive architecture that could be implemented in a robot. The aim was to implement algorithms that are based on Darwinian Neurodynamics, to perform experiments and identify future improvements.
- Two algorithm designs were implemented: co-evolution of Actors and Predictors, and evolution of Solutions (Artificial Feedforward Neural Network) that learn how to play multiple games. Experiments on both algorithms were performed using Webots robot simulator.
- The former algorithm was implemented on a simulation of a robotic dog, Aibo, the latter using simulation of a robot with two differential wheels, E-puck.
- The results of experiments on the robotic dog led to discarding both the simulation and the algorithm. The linear model of the algorithm was not suitable to predict non- linear events. Additionally, the complexity of evolving locomotion was a decisive factor in replacing the simulation with a simpler model.
- Experiments on E-puck conducted using single Game resulted in the targeted behaviour. Implementation of multiple games resulted in mediocre Solutions and led to modification of the algorithm to use Multi-Objective Optimisation.
- Further improvements that have been identified include modification of Games to increase their complexity and make them evolvable.
- GitHub repo.
Now my focus in on exams, but I will definitely continue working on this cute little yoghurt pot during summer. So many possibilities can be explored!