Visual imitation learning
---------------------------------------------------------------------------------------------------
Please send your answers to amiranas@cs.uni-freiburg.de before 19.07.2017 10:00
---------------------------------------------------------------------------------------------------
1) In what way does the human provide a high level plan to the robot? ? (1-2 sentences)

2) The authors describe the discretized action space through a tuple (p, theta, l) and treat the action selection as a classification problem. How do they avoid dealing with the huge amount of possible classes that arise when considering all possible combinations of p, theta and l? (2-3 sentences)

3) How does the robot chooses the actions during the self-supervised learning phase? (2-3 sentences)