CAREER: Planning, Execution, and Learning in Autonomous Agents

职业：自主代理的规划、执行和学习

• 批准号: 9502548
• 负责人: Manuela Veloso
• 金额: $ 14.33万
• 依托单位: Carnegie-Mellon University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 1995
• 资助国家: 美国
• 起止时间: 1995-06-15 至 1999-05-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9502548&HistoricalAwards=false
• 关键词: CAREER; Planning; Execution; Learning; Autonomous

IRI-9502548 Veloso, Manuela Carnegie-Mellon University $45,000 - 12mos CAREER: Planning, Execution, and Learning in Autonomous Robotic Agents Autonomous robotic agents need the ability to plan for and achieve high-level tasks. Current agents can achieve low-level execution goals such as object avoidance, but lack the ability to reason about and to carry out multiple interacting tasks. Integrating planning and execution by a robot is a complex task, requiring learning from prior experience to improve significantly the overall capabilities of the intelligent agent. We will explore the integration of planning, execution, and learning, where the robot's successes and failures are incorporated into a knowledge base to improve future goal-achievement performance. This project entails the investigation of a variety of system and knowledge engineering issues, including: communication between the planner and the executor; interaction between the robotic agent, the planning system, and the human user who may, at any time, request that certain tasks be performed; and gathering and interpretation of information to learn from experience. The scientific goal underlying this research is to understand the use of past experience to solve problems in complex situations affected by multiple factors with different degrees of controllability and interdependency. Even partial achievement of this goal could be of great significance to both the robotics and cognitive science or artificial intelligence research communities, bridging the representation and reasoning gap between high level planner-learner and autonomous agent executing tasks in the real world

维罗索，曼纽拉·卡内基-梅隆大学$45,000 - 12mos职业：自主机器人代理的规划、执行和学习自主机器人代理需要计划和完成高级任务的能力。当前的智能体可以实现对象回避等低级执行目标，但缺乏推理和执行多个交互任务的能力。集成机器人的规划和执行是一项复杂的任务，需要从先前的经验中学习，以显着提高智能代理的整体能力。我们将探索计划、执行和学习的整合，其中机器人的成功和失败被纳入知识库，以提高未来的目标实现绩效。这个项目需要调查各种系统和知识工程问题，包括：规划者和执行者之间的沟通；机器人代理、规划系统和人类用户之间的交互，人类用户可以随时要求执行某些任务；收集和解释信息，从经验中学习。本研究的科学目标是了解如何利用过去的经验来解决受多种不同程度可控性和相互依赖性因素影响的复杂情况下的问题。即使这一目标的部分实现，也可能对机器人和认知科学或人工智能研究界具有重要意义，弥合了在现实世界中执行任务的高级计划-学习者和自主代理之间的表示和推理差距