Risk Averse Bayesian Reward Learning for Autonomous Navigation from Human Demonstration

Christian Ellis; Maggie Wigness; John G. Rogers III; Craig Lennon; Lance Fiondella

doi:10.1109/iros51168.2021.9635835

All publications

Conference paper · 2021

Learning from Demonstration & Reinforcement Learning

Risk Averse Bayesian Reward Learning for Autonomous Navigation from Human Demonstration

Christian Ellis, Maggie Wigness, John G. Rogers III, Craig Lennon, Lance Fiondella

2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 8928-8935, 2021

PDF DOI: 10.1109/iros51168.2021.9635835 arXiv Publisher

Three terrain grid maps: a training map, a test map that adds water, and planned paths where Bayesian reward learning avoids water and MaxEnt IRL crosses it. — Fig. 1. Simplified illustration to demonstrate the impact of distributional shift. Left: Training environment. Middle: Test environment. Right: Planned trajectories in the testing environment for various learning methods.

Abstract

The study presents a Bayesian approach to reward learning for autonomous navigation systems, emphasizing risk aversion and learning from human demonstrations to improve navigation safety and efficiency.

Cite

@inproceedings{ellis2021risk,
  title = {{Risk Averse Bayesian Reward Learning for Autonomous Navigation from Human Demonstration}},
  author = {Christian Ellis and Maggie Wigness and John G. Rogers III and Craig Lennon and Lance Fiondella},
  year = {2021},
  booktitle = {2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  pages = {8928--8935},
  doi = {10.1109/iros51168.2021.9635835}
}