RI: Small: Cooperative Planning and Learning via Scalable and Learnable Multi-Agent Commitments

承诺的概念起源于人类社会，指的是决策者或代理人对其未来决策后果的各个方面做出可信和长期的承诺，从而促进与其他代理人的合作。因此，工程承诺是一个很有前途的框架，以实现合作的人工智能（AI），装备一组自主代理的规划和学习能力，以最大限度地提高他们的联合效用。该研究项目旨在启动一种范式转变，通过将其扩展到合作AI中复杂性的各个维度，将承诺的概念充分发挥其潜力，开发新的方法，这些方法有望对现实世界和大规模合作AI应用产生重大而积极的影响。该项目整合了一系列教育举措，在PI的课程中发挥关键作用，招募和培训来自代表性不足背景的本科生，以及计划让高中生和初级研究人员参与的广泛活动。推力1重新设计了现有的基于承诺的分布式合作规划方法，具有预定义的概率承诺参数，通过开发新的算法和分析，在规划的约束和不确定性，近似线性规划，和鲁棒规划，解决长决策范围和高维感知和行动;推力2开发和评估了一种新的分布式合作学习方法与紧急承诺参数化，它结合了多代理承诺框架和深度强化学习的最佳效果，以解决上述所有复杂性问题。该研究的成功将大大提高基于承诺的规划和学习在大规模和复杂的合作AI系统中的适用性。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而被认为值得支持。

Stemming from human societies, the notion of commitment refers to a decision maker, or agent, making credible and prolonged promises about various aspects of the consequences of its future decisions, thus facilitating cooperation with other agents. Engineering commitments is therefore a promising framework to achieve cooperative artificial intelligence (AI) that equips a group of autonomous agents with the capability of planning and learning to maximize their joint utility. This research project seeks to initiate a paradigm shift that brings the notion of commitment to its full potential by scaling it to various dimensions of complexity in cooperative AI, developing novel methods that promise to significantly and positively impact real-world and large-scale cooperative AI applications. This project integrates an array of education initiatives, playing key roles in PI's classes, the recruitment and training of undergraduate students from underrepresented backgrounds, and extensive activities planned to involve high school students and junior researchers.This research consists of two cohesive thrusts: Thrust 1 redesigns an existing approach for commitment-based distributed cooperative planning with a predefined parameterization for probabilistic commitments, by developing novel algorithms and analyses in planning under constraints and uncertainty, approximate linear programming, and robust planning that address long decision horizon and high-dimensional perception and action; Thrust 2 develops and evaluates a novel approach for distributed cooperative learning with emergent commitment parameterization, which combines the best from the framework of multi-agent commitments and deep reinforcement learning to address all aforementioned dimensions of complexity. Success of the proposed research is expected to significantly increase the applicability of commitment-based planning and learning for large-scale and complex cooperative AI systems.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.