Flow: A Modular Learning Framework for Mixed Autonomy Traffic
The rapid development of autonomous vehicles (AVs) holds vast potential for transportation systems through improved safety, efficiency, and access to mobility. However, the progression of these impacts, as AVs are adopted, is not well understood. Numerous technical challenges arise from the goal...
| Main Authors: | , , , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
Institute of Electrical and Electronics Engineers (IEEE)
2023
|
| Online Access: | https://hdl.handle.net/1721.1/148679 |
| _version_ | 1826188416250806272 |
|---|---|
| author | Wu, Cathy Kreidieh, Abdul Rahman Parvate, Kanaad Vinitsky, Eugene Bayen, Alexandre M |
| author2 | Massachusetts Institute of Technology. Department of Civil and Environmental Engineering |
| author_facet | Massachusetts Institute of Technology. Department of Civil and Environmental Engineering Wu, Cathy Kreidieh, Abdul Rahman Parvate, Kanaad Vinitsky, Eugene Bayen, Alexandre M |
| author_sort | Wu, Cathy |
| collection | MIT |
| description | The rapid development of autonomous vehicles (AVs) holds vast potential for
transportation systems through improved safety, efficiency, and access to
mobility. However, the progression of these impacts, as AVs are adopted, is not
well understood. Numerous technical challenges arise from the goal of analyzing
the partial adoption of autonomy: partial control and observation,
multi-vehicle interactions, and the sheer variety of scenarios represented by
real-world networks. To shed light into near-term AV impacts, this article
studies the suitability of deep reinforcement learning (RL) for overcoming
these challenges in a low AV-adoption regime. A modular learning framework is
presented, which leverages deep RL to address complex traffic dynamics. Modules
are composed to capture common traffic phenomena (stop-and-go traffic jams,
lane changing, intersections). Learned control laws are found to improve upon
human driving performance, in terms of system-level velocity, by up to 57% with
only 4-7% adoption of AVs. Furthermore, in single-lane traffic, a small neural
network control law with only local observation is found to eliminate
stop-and-go traffic - surpassing all known model-based controllers to achieve
near-optimal performance - and generalize to out-of-distribution traffic
densities. |
| first_indexed | 2024-09-23T07:59:17Z |
| format | Article |
| id | mit-1721.1/148679 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T07:59:17Z |
| publishDate | 2023 |
| publisher | Institute of Electrical and Electronics Engineers (IEEE) |
| record_format | dspace |
| spelling | mit-1721.1/1486792023-03-24T03:54:28Z Flow: A Modular Learning Framework for Mixed Autonomy Traffic Wu, Cathy Kreidieh, Abdul Rahman Parvate, Kanaad Vinitsky, Eugene Bayen, Alexandre M Massachusetts Institute of Technology. Department of Civil and Environmental Engineering The rapid development of autonomous vehicles (AVs) holds vast potential for transportation systems through improved safety, efficiency, and access to mobility. However, the progression of these impacts, as AVs are adopted, is not well understood. Numerous technical challenges arise from the goal of analyzing the partial adoption of autonomy: partial control and observation, multi-vehicle interactions, and the sheer variety of scenarios represented by real-world networks. To shed light into near-term AV impacts, this article studies the suitability of deep reinforcement learning (RL) for overcoming these challenges in a low AV-adoption regime. A modular learning framework is presented, which leverages deep RL to address complex traffic dynamics. Modules are composed to capture common traffic phenomena (stop-and-go traffic jams, lane changing, intersections). Learned control laws are found to improve upon human driving performance, in terms of system-level velocity, by up to 57% with only 4-7% adoption of AVs. Furthermore, in single-lane traffic, a small neural network control law with only local observation is found to eliminate stop-and-go traffic - surpassing all known model-based controllers to achieve near-optimal performance - and generalize to out-of-distribution traffic densities. 2023-03-23T16:44:16Z 2023-03-23T16:44:16Z 2022 2023-03-23T15:53:46Z Article http://purl.org/eprint/type/ConferencePaper https://hdl.handle.net/1721.1/148679 Wu, Cathy, Kreidieh, Abdul Rahman, Parvate, Kanaad, Vinitsky, Eugene and Bayen, Alexandre M. 2022. "Flow: A Modular Learning Framework for Mixed Autonomy Traffic." IEEE Transactions on Robotics, 38 (2). en 10.1109/TRO.2021.3087314 IEEE Transactions on Robotics Creative Commons Attribution-Noncommercial-Share Alike https://creativecommons.org/licenses/by-nc-sa/4.0/ application/pdf Institute of Electrical and Electronics Engineers (IEEE) arXiv |
| spellingShingle | Wu, Cathy Kreidieh, Abdul Rahman Parvate, Kanaad Vinitsky, Eugene Bayen, Alexandre M Flow: A Modular Learning Framework for Mixed Autonomy Traffic |
| title | Flow: A Modular Learning Framework for Mixed Autonomy Traffic |
| title_full | Flow: A Modular Learning Framework for Mixed Autonomy Traffic |
| title_fullStr | Flow: A Modular Learning Framework for Mixed Autonomy Traffic |
| title_full_unstemmed | Flow: A Modular Learning Framework for Mixed Autonomy Traffic |
| title_short | Flow: A Modular Learning Framework for Mixed Autonomy Traffic |
| title_sort | flow a modular learning framework for mixed autonomy traffic |
| url | https://hdl.handle.net/1721.1/148679 |
| work_keys_str_mv | AT wucathy flowamodularlearningframeworkformixedautonomytraffic AT kreidiehabdulrahman flowamodularlearningframeworkformixedautonomytraffic AT parvatekanaad flowamodularlearningframeworkformixedautonomytraffic AT vinitskyeugene flowamodularlearningframeworkformixedautonomytraffic AT bayenalexandrem flowamodularlearningframeworkformixedautonomytraffic |