项目名称: 基于低维连续表示的启发式智能规划技术研究
项目编号: No.61502227
项目类型: 青年科学基金项目
立项/批准年度: 2016
项目学科: 其他
项目作者: 张雷
作者单位: 南京大学
项目金额: 21万元
中文摘要: 启发式智能规划是智能规划和人工智能领域一个广为研究的重要问题。现有的启发式智能规划的研究通过对规划任务的表示进行分析提取出领域无关的启发式函数,从而加速规划求解过程,取得了可观的进展。但是大多数研究都是使用离散高维向量来表示规划任务,高维的规划状态空间导致了维数灾难,限制了规划推理算法效率的提高。本项目在分析现有的启发式智能规划表示的基础上,提出利用低维连续向量来表示规划状态。同时,在此基础上将启发式函数抽取,landmark知识发现,规划解抽取等启发式智能规划推理问题建模为低维连续向量空间中的推理问题。这不仅提供了一种利用低维表示学习进行启发式智能规划表示的新框架,还可以提供一种启发式函数抽取,landmark知识发现和规划解抽取的新思路,促进智能规划领域的发展。
中文关键词: 智能规划;规划表示;启发式函数设计;地标规划
英文摘要: Heuristic search planning is an widely studied research topic in Artificial Intelligence(AI) and AI planning. Existing heuristic based planners speed up planning processes by using domain independent heuristic functions that are automatically derived from planning task representations. Despite the success of this approach, most research represent planning tasks using discrete state spaces that are very high dimensional, which makes it very hard to derive efficient heuristic functions and dong inference. Based on analysis of existing planning representations, the project models planning tasks using continuous vectors whose dimensions are relatively low, and models planning inference tasks as heuristic function extraction, landmark extraction and plan extractions as inference tasks in the low dimensional continuous vector spaces. This can not only provide a new way to representation planning tasks using low dimensional continuous vectors, but also offer new frameworks for extracting heuristic functions, discovering landmarks and obtaining planning solutions, which will promote the development of AI planning.
英文关键词: AI Planning;PLanning Representation;Heuristic Function Design;Landmark Planning