Learning Neuro-Symbolic Relational Transition Models for Bilevel Planning
Learning Neuro-Symbolic Relational Transition Models for Bilevel Planning
复制标题
DOI：
10.1109/iros47612.2022.9981440
复制
发表时间：
2021-05
期刊：
影响因子：
--
通讯作者：
Rohan Chitnis;Tom Silver;J. Tenenbaum;Tomas Lozano-Perez;L. Kaelbling
中科院分区：
文献类型：
--
作者：
Rohan Chitnis;Tom Silver;J. Tenenbaum;Tomas Lozano-Perez;L. Kaelbling
在机器人领域中，学习和规划由连续的状态空间、连续的动作空间和较长的任务视界组成。在这项工作中，我们用神经-符号关系转换模型(NSRT)来解决这些挑战，NSRT是一类新的模型，它学习起来数据高效，与强大的机器人规划方法兼容，并且可以在对象上泛化。NSRT同时具有符号和神经组件，从而实现了双层规划方案，其中外环中的符号AI规划指导内环中的神经模型的连续规划。在四个机器人规划领域的实验表明，NSRT可以非常高效地学习数据，然后用于新任务的快速规划，这些任务需要多达60个动作，涉及的对象比训练中看到的多得多。
In robotic domains, learning and planning are complicated by continuous state spaces, continuous action spaces, and long task horizons. In this work, we address these challenges with Neuro-Symbolic Relational Transition Models (NSRTs), a novel class of models that are data-efficient to learn, compatible with powerful robotic planning methods, and generalizable over objects. NSRTs have both symbolic and neural components, enabling a bilevel planning scheme where symbolic AI planning in an outer loop guides continuous planning with neural models in an inner loop. Experiments in four robotic planning domains show that NSRTs can be learned very data-efficiently, and then used for fast planning in new tasks that require up to 60 actions and involve many more objects than were seen during training.