Collaborative Research: Interaction-aware Planning and Control for Robotic Navigation in the Crowd

该项目旨在使机器人能够在拥挤、动态的环境中导航，如城市街道和忙碌的人行道。例如，考虑几个小型地面交付机器人，它们必须导航到特定的目标位置，同时避开多个行人。目前，决策算法遵循“预测然后计划”的方法，其中机器人预测场景中代理的未来运动，并随后计划避免机动。然而，在现实中，每个智能体当前的决策会影响其他智能体未来的观察和决策问题。这种通过时间的最优规划的耦合很自然地表达在动态博弈论的形式主义中;然而，不幸的是，一般动态博弈的实用和有效的求解方法长期以来一直难以捉摸。该项目开发了理论和算法技术，以解决一些潜在的挑战，并将支持多个博士生的跨机构辅导，本科课程材料的开发，并推广到当地代表性不足的社区。该项目的具体目标是三重。第一个目标是算法，旨在构建新的算法，以找到近似平衡的解决方案，在几个常见的类的动态游戏模型的不同模式的人机交互。由于这些算法解决了机器人导航问题，它们还必须能够嵌入式机载实现。该项目的第二个目标是解决“逆”问题：在人群中的最佳规划取决于对人类目标的预知。鉴于现有的技术推断代理的目标孤立，该项目的目的是获得新的方法的战略耦合设置。第三个也是最后一个目标是通过计算并行化和分散化来加速多机器人人群场景中的交互感知规划。这些算法将在人群导航的环境中使用四轴飞行器和地面移动的机器人与人类受试者进行广泛的评估。该项目得到了机器人项目跨董事会基础研究的支持，由工程局（ENG）和计算机与信息科学与工程局（CISE）共同管理和资助该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This project aims to enable robot navigation in crowded, dynamic environments such as urban streets and busy walkways. For example, consider several small ground delivery robots which must navigate to specific goal positions while avoiding multiple pedestrians. Currently, decision-making algorithms follow a "predict then plan" approach, in which robots predict the future motion of agents in a scene and subsequently plan avoidance maneuvers. In reality, however, each agent's current decision affects the future observations and decision problems faced by others. This coupling of optimal planning through time is naturally expressed in the formalism of dynamic game theory; unfortunately, however, practical and efficient solution methods for general dynamic games have long been elusive. This project develops theoretical and algorithmic techniques to address some of the underlying challenges, and will also support cross-institution mentoring of multiple PhD students, development of undergraduate course material, and outreach to local underrepresented communities.The specific goals of this project are threefold. The first goal is algorithmic, and aims to construct new algorithms to find approximate equilibrium solutions in several common classes of dynamic games which model distinct modes of human-robot interaction. As these algorithms solve robotic navigation problems, they must also be amenable to embedded, onboard implementation. The second goal of this project addresses the "inverse" problem: optimal planning in a crowd depends upon foreknowledge of humans' objectives. Whereas existing techniques infer agents' objectives in isolation, this project aims to derive novel methods for the strategically-coupled setting. The third and final goal is to accelerate interaction-aware planning in multi-robot, crowd scenarios via computational parallelization and decentralization. The algorithms will be extensively evaluated with human subjects in the setting of crowd navigation, using quadcopters and ground mobile robots.This project is supported by the cross-directorate Foundational Research in Robotics program, jointly managed and funded by the Directorates for Engineering (ENG) and Computer and Information Science and Engineering (CISE).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.