Collaborative Research: Mixed-Autonomy Traffic Networks: Routing Games and Learning Human Choice Models

合作研究：混合自主交通网络：路由博弈和学习人类选择模型

• 批准号: 1953032
• 负责人: Dorsa Sadigh
• 金额: $ 18万
• 依托单位: Stanford University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-08-15 至 2023-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1953032&HistoricalAwards=false
• 关键词: Collaborative; Research; Mixed; Autonomy; Traffic

Autonomous and connected vehicles are soon becoming a significant part of roads normally used by human drivers. Such vehicles hold the promise of safer streets, better fuel efficiency, more flexibility in tailoring to specific drivers’ needs, and time savings. However, the appearance of autonomous vehicles driving on roads shared by human-driven cars introduce many interesting and timely challenges. The goal of this proposal is to study (i) traffic networks with mixed autonomy where a fraction of cars are autonomous and the rest are human-driven, and (ii) how humans choose their routes in a traffic network given different options of autonomous service and prices. By studying models of humans’ choices and investigating the characterizations of traffic flow in networks with mixed autonomy, the project develops routing policies to lead the network to an efficient equilibrium with low average latency.This proposal aims to study routing games and human choice models for traffic networks with mixed autonomy. Many studies have shown that mobility can be enhanced in traffic networks such as freeways or signalized intersections when all cars are autonomous; however, such improvement is far from clear for a network with mixed autonomy. The goal of this project is to study the game theory of mixed-autonomy traffic networks and control the autonomous cars’ routing decisions such that the system reaches an optimum equilibrium. Moreover, a novel approach in learning human choices of prices in autonomous transportation services versus latency, or travel time, is developed. Finally, using the well-known fundamental diagram of traffic and cell-transmission model, a dynamic mixed-autonomy traffic model is introduced. Using this dynamic model, we will leverage tools from reinforcement learning to route autonomous cars dynamically and optimally. The proposed research considers both theoretical study of routing games as well as implementation of the developed algorithms in traffic simulators, in particular simulation of Urban Mobility (SUMO). The learned human choice models will also be validated through human subject studies.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

自动驾驶和联网车辆很快就会成为人类驾驶员通常使用的道路的重要组成部分，此类车辆有望提供更安全的街道、更好的燃油效率、更灵活地满足特定驾驶员的需求以及节省时间。在人类驾驶汽车共享的道路上行驶的自动驾驶汽车带来了许多有趣且及时的挑战，该提案的目标是研究（i）具有混合自动驾驶的交通网络，其中一部分汽车是自动驾驶的，其余是人类驾驶的，并且(ii) 人类如何在给定的交通网络中选择路线通过研究人类选择模型并研究混合自治网络中流量的特征，该项目开发路由策略以引导网络达到低平均延迟的有效平衡。该提案旨在研究具有混合自主性的交通网络的路线博弈和人类选择模型，表明当所有汽车都是自动驾驶时，高速公路或信号交叉口等交通网络的流动性可以得到增强，但是对于具有混合自主性的网络来说，这种改进还远未明确。该项目的目标是研究混合自治。此外，还开发了一种学习人类对自主交通服务价格与延迟或旅行时间的选择的新方法。最后，使用众所周知的交通和单元传输模型的基本图，引入了动态混合自主交通模型，我们将利用强化学习的工具来动态优化自动驾驶汽车的路线。既考虑了路由博弈的理论研究，也考虑了在交通模拟器中实施所开发的算法，特别是城市交通 (SUMO) 的模拟，还将通过人类主题研究来验证所学习的人类选择模型。该奖项反映了 NSF 的法定使命，并通过使用该奖项的评估被认为值得支持。基金会的智力价值和更广泛的影响审查标准。

项目成果

Emergent Prosociality in Multi-Agent Games Through Gifting

通过礼物在多智能体游戏中出现的亲社会性

• DOI: 10.24963/ijcai.2021/61
• 发表时间: 2021-05-13
• 期刊: ArXiv
• 作者: Woodrow Z. Wang;M. Beliaev;Erdem Biyik;Daniel A. Lazar;Ramtin Pedarsani;Dorsa Sadigh
• 通讯作者: Dorsa Sadigh

Partner-Aware Algorithms in Decentralized Cooperative Bandit Teams

去中心化合作强盗团队中的合作伙伴感知算法

• 发表时间: 2022-01
• 期刊: Proceedings of the 36th AAAI Conference on Artificial Intelligence
• 作者: Erdem Biyik; Anusha Lalitha
• 通讯作者: Anusha Lalitha

Cooperative Autonomous Vehicles that Sympathize with Humans

与人类产生共鸣的协作式自动驾驶汽车

• 发表时间: 2021-01
• 期刊: Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS
• 作者: Behrad Toghi; Rodolfo Valiente
• 通讯作者: Rodolfo Valiente

PantheonRL: A MARL Library for Dynamic Training Interactions

PantheonRL：用于动态训练交互的 MARL 库

• DOI: 10.1609/aaai.v36i11.21734
• 发表时间: 2021-12-13
• 期刊: Microbiology
• 影响因子: 1.5
• 作者: Bidipta Sarkar;Aditi Talati;Andy Shih;Dorsa Sadigh
• 通讯作者: Dorsa Sadigh

Learning How to Dynamically Route Autonomous Vehicles on Shared Roads

学习如何在共享道路上动态规划自动驾驶车辆的路线

• DOI: 10.1016/j.trc.2021.103258
• 发表时间: 2019-09-09
• 期刊: ArXiv
• 作者: Daniel A. Lazar;Erdem Biyik;Dorsa Sadigh;Ramtin Pedarsani
• 通讯作者: Ramtin Pedarsani