Things to try

In this chapter, we only started playing with MiniWoB by touching upon the six easiest environments from the full set of 80 problems, so there is plenty of uncharted territory ahead. If you want to practice, there are several items you can experiment with:

• Testing the robustness of demonstrations to noisy clicks.
• Implementing training of the value head of A3C based on demonstration data.
• Implementing more sophisticated mouse control, like move mouse N pixels left/right/top/bottom.
• Using some pretrained optical character recognition (OCR) network (or training your own!) to extract text information from the observations.
• Taking other problems and trying to solve them. There are some quite tricky and fun problems, like sort items using drag-n-drop or repeat the pattern using checkboxes.
• Checking MiniWoB++ (https://stanfordnlp.github.io/miniwob-plusplus/) from the Stanford NLP Group. It will require learning and writing new wrappers; as mentioned...