CaltechAUTHORS
  A Caltech Library Service

Meta Inverse Reinforcement Learning via Maximum Reward Sharing for Human Motion Analysis

Li, Kun and Burdick, Joel W. (2017) Meta Inverse Reinforcement Learning via Maximum Reward Sharing for Human Motion Analysis. In: Workshop on Meta-Learning (MetaLearn 2017), 9 December 2017, Long Beach, CA. https://resolver.caltech.edu/CaltechAUTHORS:20190410-120626845

[img] PDF - Published Version
See Usage Policy.

131Kb

Use this Persistent URL to link to this item: https://resolver.caltech.edu/CaltechAUTHORS:20190410-120626845

Abstract

This work handles the inverse reinforcement learning (IRL) problem where only a small number of demonstrations are available from a demonstrator for each high-dimensional task, insufficient to estimate an accurate reward function. Observing that each demonstrator has an inherent reward for each state and the task-specific behaviors mainly depend on a small number of key states, we propose a meta IRL algorithm that first models the reward function for each task as a distribution conditioned on a baseline reward function shared by all tasks and dependent only on the demonstrator, and then finds the most likely reward function in the distribution that explains the task-specific behaviors. We test the method in a simulated environment on path planning tasks with limited demonstrations, and show that the accuracy of the learned reward function is significantly improved.


Item Type:Conference or Workshop Item (Paper)
Related URLs:
URLURL TypeDescription
http://metalearning.ml/2017/papers/metalearn17_li.pdfPublisherArticle
http://arxiv.org/abs/1710.03592arXivArticle
Record Number:CaltechAUTHORS:20190410-120626845
Persistent URL:https://resolver.caltech.edu/CaltechAUTHORS:20190410-120626845
Usage Policy:No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:94631
Collection:CaltechAUTHORS
Deposited By: George Porter
Deposited On:11 Apr 2019 14:50
Last Modified:03 Oct 2019 21:05

Repository Staff Only: item control page