Building Digital Twins via GPU-Accelerated Human-Regularized Reinforcement Learning

Vinitsky, Eugene; Cornelisse, Daphne

i

Building Digital Twins via GPU-Accelerated Human-Regularized Reinforcement Learning

2024-09-01
By Vinitsky, Eugene ; Cornelisse, Daphne

File Language:

English

Details

Creators:

Vinitsky, Eugene ; Cornelisse, Daphne
Corporate Creators:

Connected Communities for Smart Mobility Toward Accessible and Resilient Transportation for Equitably Reducing Congestion (C2SMARTER) Tier-1 University Transportation Center (UTC)
Corporate Publisher:

Connected Communities for Smart Mobility Toward Accessible and Resilient Transportation for Equitably Reducing Congestion (C2SMARTER) Tier-1 University Transportation Center (UTC)
Corporate Contributors:

United States. Department of Transportation. Federal Highway Administration. Office of Research, Development, and Technology ; United States. Department of Transportation. University Transportation Centers (UTC) Program ; United States. Department of Transportation. Office of the Assistant Secretary for Research and Technology
Subject/TRT Terms:
Resource Type:

Tech Report
Geographical Coverage:

United States
Edition:

Final report 09/30/2023-09/30/2024
Description:

We have carried out a two-component study investigating how to build high-performing human driver models. First, we investigated how to combine large-scale datasets from autonomous vehicle companies with reinforcement learning methods for training agents. We showed that by adding a supervised learning loss atop the model training, we could build models that were human-like without reducing the agent performance. However, we were bottlenecked by simulator speed. We then designed anew simulator, Gurdave, that can run thousands of simultaneous simulations of urban environments containing drivers, pedestrians, and cyclists. This simulator is an order of magnitude faster than existing open-source simulators. We demonstrate that in this simulator, we can reproduce our prior agent training results between one and two orders of magnitude faster.
Format:

PDF
Funding:

69A3552348326
Download URL:

https://rosap.ntl.bts.gov/view/dot/86354/dot_86354_DS1.pdf
File Type:

[PDF - 1.70 MB]
Collection(s):

Artificial Intelligence Deployment Center of Excellence (AI COE) ; University Transportation Centers
Main Document Checksum:

urn:sha-512:d5b76881a0f047aedf9551a87783a8fc6b91ae4ff51d9e6d850a29980f6fe1b9d153524a42b438ed47539bcdaced4f2c91767e89e974d7c76cd0c9f06007e948

Building Digital Twins via GPU-Accelerated Human-Regularized Reinforcement Learning [supporting dataset]

Dataset

We have carried out a two-component study investigating how to build high-performing human driver models. First, we investigated how to combine large- ...

Vinitsky, Eugene ;

Cornelisse, Daphne
Building Reliable Sim Driving Agents by Scaling Self-Play

Simulation agents are essential for designing and testing systems that interact with humans, such as autonomous vehicles (AVs). These agents serve var ...

Cornelisse, Daphne ;

Pandya, Aarav

File Language:

English

ON THIS PAGE

Details Related Documents

Found an issue?

Ask a Librarian at:

https://transportation.libanswers.com/

ROSA P serves as an archival repository of USDOT-published products including scientific findings, journal articles, guidelines, recommendations, or other information authored or co-authored by USDOT or funded partners. As a repository, ROSA P retains documents in their original published format to ensure public access to scientific information.