#93 Incorporating Uncertainty for Reinforcement Learning of Agent Policies
- Principal Investigator
- Daniel Lee
- Status
- Completed
- Start Date
- Jan. 1, 2018
- End Date
- Dec. 28, 2018
- Project Type
- Research Advanced
- Grant Program
- FAST Act - Mobility National (2016 - 2022)
- Grant Cycle
- Mobility21 - University of Pennsylvania
- Visibility
- Public
Abstract
This project investigates modeling uncertainty for reinforcement learning of agent policies
in Markov Decision Processes. The idea is to approximate the posterior over beliefs of
the Q values with experience. The beliefs are combined in real-time using Assumed
Density Filtering (ADF) with Temporal Difference Learning. Applications of this
reinforcement learning algorithms include learning autonomous behaviors and
modeling driver intentions.
Description
Timeline
Strategic Description / RD&T
Deployment Plan
Expected Outcomes/Impacts
Expected Outputs
TRID
Individuals Involved
| Name | Affiliation | Role | Position | |
|---|---|---|---|---|
| heejinj@seas.upenn.edu | Jeong, Heejin | University of Pennsylvania | Other | Student - PhD |
| ddlee@seas.upenn.edu | Lee, Daniel | University of Pennsylvania | PI | Faculty - Tenured |
Budget
Amount of UTC Funds Awarded
$Total Project Budget (from all funding sources)
$Documents
| Type | Name | Uploaded |
|---|---|---|
| Publication | adfq_aaai_v1.pdf | Sept. 30, 2018, 12:09 p.m. |
| Presentation | Bayesian Q-learning with Assumed Density Filtering | March 30, 2018, 6:57 a.m. |
| Progress Report | 93_Progress_Report_2018-03-30 | March 30, 2018, 6:58 a.m. |
| Progress Report | 93_Progress_Report_2018-09-30 | Sept. 30, 2018, 12:09 p.m. |
Match Sources
No match sources!
Partners
No partners!