Extrapolating Beyond Suboptimal Demonstrations via Inverse Reinforcement Learning from Observations
share
A critical flaw of existing inverse reinforcement learning (IRL) methods is their inability to significantly outperform the demonstrator. This is a consequence of the general reliance of IRL algorithms upon some form of mimicry, such as feature-count matching, rather than inferring the underlying intentions of the demonstrator that may have been poorly executed in practice. In this paper, we introduce a novel reward learning from observation algorithm, Trajectory-ranked Reward EXtrapolation (T-REX), that extrapolates beyond a set of (approximately) ranked demonstrations in order to infer high-quality reward functions from a set of potentially poor demonstrations. When combined with deep reinforcement learning, we show that this approach can achieve performance that is more than an order of magnitude better than the best-performing demonstration, on multiple Atari and MuJoCo benchmark tasks. In contrast, prior state-of-the-art imitation learning and IRL methods fail to perform better than the demonstrator and often have performance that is orders of magnitude worse than T-REX. Finally, we demonstrate that T-REX is robust to modest amounts of ranking noise and can accurately extrapolate intention by simply watching a learner noisily improve at a task over time.
READ FULL TEXT