Feature Construction for Inverse Reinforcement Learning

By Sergey Levine, Zoran Popovic, Vladlen Koltun. 2010

126 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

A Reinforcement Learning Approach to Weaning of Mechanical Ventilation in Intensive Care Units (2017, methodology): “Possible directions for future work also include increasing the sophistication of the state space, for example by handling long term effects more explicitly using second-order statistics of vitals, applying techniques in inverse reinforcement learning to feature engineering (as in Levine et al. [201”
Concrete Problems for Autonomous Vehicle Safety: Advantages of Bayesian Deep Learning (2017, methodology): “In this method we learn a reward function given features from a particular scene, aiming to generalise well to new scenes [Levine et al., 2010].”
Maximum Entropy Deep Inverse Reinforcement Learning (2015, background): “To overcome the inherent limitations of linear models, (Choi & Kim, 2013) and (Levine et al., 2010) extend this approach to a limited set of non-linear rewards by learning a set of composites of logical conjunctions of atomic features.”
Nonlinear Inverse Reinforcement Learning with Gaussian Processes (2011, methodology): “While the performance of MaxEnt and FIRL also improved with additional examples, they were consistently outperformed by GPIRL.”
Learning strategies in table tennis using inverse reinforcement learning (2014, methodology): “An automatic approach for choosing these reward features was suggested by Levine et al. (2010). Here, it was suggested to construct the features from a logical combinations of components that are the most relevant to the task.”
Preference-Based Policy Learning (2011, unknown): “Compared to the discriminant policy representation built from the set of policy demonstrations [17], the main two differences are that BvR is independent of the policy return estimate, and that it is built online as new policies explore new regions of the sensori-motor space.”
Continuous Inverse Optimal Control with Locally Optimal Examples (2012, methodology): “We present two variants of our algorithm that learn the reward either as a linear combination of the provided features, as is common in prior work, or as a nonlinear function of the features, as in a number of recent methods (Ratliff et al., 2009; Levine et al., 2010; 2011).”
A Survey of Inverse Reinforcement Learning: Challenges, Methods and Progress (2018, methodology): “Some methods showed an improvement in this context [42, 39, 34].”
Feature Expansive Reward Learning: Rethinking Human Input (2020, background): “This limitation can 1Prior work proposed tackling this issue by iterative boosting (Ratliff et al., 2007) or constructing binary features (Levine et al., 2010; Choi & Kim, 2013).”
Inverse Reinforcement Learning algorithms and features for robot navigation in crowds: An experimental comparison (2014, unknown): “[21]).”