A Unified Framework for Inferring Rewards from Diverse Types of Human Feedback
Abstract
It is often difficult to hand-specify what the correct reward function is for a task, so researchers have instead aimed to enable robots to learn reward functions from human behavior or feedback. The types of behavior interpreted as evidence of the reward function have expanded greatly in recent years. We ve gone from demonstrations, to comparisons, to reading into the information leaked when the human is pushing the robot away or turning it off. And surely, there is more to come. This project aims to develop a unified framework for human feedback. Our scientific goal is to better understand the diversity of human feedback. Our practical goal is to enable robots to combine and actively query for different known feedback types, as well as to enable researchers to more readily identify and algorithmically interpret new sources of information about reward functions.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Oct 19, 2020
- Source ID
- N000142012736
Entities
People
- Anca Dragan
Organizations
- Office of Naval Research
- United States Navy
- University of California Regents