Learning from Demonstrations and Human Evaluative Feedbacks: Handling Sparsity and Imperfection Using Inverse Reinforcement Learning Approach

Table 1

Strengths and weaknesses of and approaches.

	Learning from evaluative feedbacks ()	Learning from demonstrations ()
Strengths	(i) Simplicity, where “teachers” teach “learners” without needing detailed knowledge of how to perform the task themselves. It is enough for the teacher to evaluate the outcome; (ii) Feedback is not affected by correspondence problem between the learner and the teacher (the physical differences)	(i) Effective transfer learning techniques, where the learner generalizes from teacher’s demonstration to state-action mapping in whole space; (ii) Speeds up the learning process and reduces the regret, because the teacher is providing the correct action directly; (iii) Decreases the learner exploration to get the correct action
Weaknesses	(i) Learning process is slow and needs a large number of teacher’s feedbacks. So, it is a boring job; (ii) The learner behaves randomly at early learning trials; (iii) Inconsistency and errors during providing feedback	(i) The teacher must have a clear policy in her mind to provide demonstrations and must be an expert in doing the task; (ii) Demonstrations cannot be easily available in some situations due to danger or low possibility of occurrence; (iii) Different physical embodiment and perception between the teacher and the learner (correspondence problem)