A fast hybrid reinforcement learning framework with human corrective feedback

Carlos Celemin*, Javier Ruiz-del-Solar, Jens Kober

*Corresponding author for this work

Research output: Contribution to journalArticleScientificpeer-review

114 Downloads (Pure)

Abstract

Reinforcement Learning agents can be supported by feedback from human teachers in the learning loop that guides the learning process. In this work we propose two hybrid strategies of Policy Search Reinforcement Learning and Interactive Machine Learning that benefit from both sources of information, the cost function and the human corrective feedback, for accelerating the convergence and improving the final performance of the learning process. Experiments with simulated and real systems of balancing tasks and a 3 DoF robot arm validate the advantages of the proposed learning strategies: (i) they speed up the convergence of the learning process between 3 and 30 times, saving considerable time during the agent adaptation, and (ii) they allow including non-expert feedback because they have low sensibility to erroneous human advice.

Original languageEnglish
Pages (from-to)1173-1186
JournalAutonomous Robots
Volume43 (2019)
Issue number5
DOIs
Publication statusPublished - 2018

Keywords

  • Interactive machine learning
  • Learning from demonstration
  • Policy search
  • Reinforcement learning

Fingerprint

Dive into the research topics of 'A fast hybrid reinforcement learning framework with human corrective feedback'. Together they form a unique fingerprint.

Cite this