Automated Planning involves taking actions to reach the goal. The problem is that most of the practical situations have lots of actions involved, and given a situation, some of these actions might be unnecessary for computing the plan. Therefore planning agents take a lot of time for computing plans because they might be considering redundant actions while planning.
The first part of this project is to use Machine Learning algorithms to learn which actions can be useful for computing the plan. Therefore the idea is to train our learning algorithms on many such instances, and given a planning problem, output a probability distribution over all the actions (which indicates the relevance of each action).
We'll be applying following learning algorithms for getting the probability distribution over actions:
- Bayes Classifier
- Artificial Neural Networks
The second part of the project is to improve AGGLEditor. For this part, we’ll resolve bugs that are present in the current version of the editor. Documentation and tutorials for AGGLEditor will also be added.