Learning Neuro-Symbolic Relational Transition Models for Bilevel Planning
This work addresses robotic planning problems by combining symbolic and neural AI for improved efficiency and generalization, representing a novel method for a known bottleneck.
The paper tackles the challenges of learning and planning in robotic domains with continuous state and action spaces by introducing Neuro-Symbolic Relational Transition Models (NSRTs), which enable bilevel planning and achieve data-efficient learning with only tens or hundreds of training episodes, allowing fast planning in new tasks requiring up to 60 actions and involving more objects than seen during training.
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 after only tens or hundreds of training episodes, 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. Video: https://tinyurl.com/chitnis-nsrts