PDSketch: Integrated Planning Domain Programming and Learning

TL;DR

PDSketch introduces a flexible domain definition language that leverages environmental structure to improve planning efficiency and generalization in robotic models, combining domain programming with neural network learning.

Contribution

It presents PDSketch, a novel language for defining transition models with structural information, enabling automatic heuristic generation for faster planning.

Findings

Improved model generalization and data efficiency.

Faster planning for new goals using learned heuristics.

Effective exploitation of environmental structure in models.

Abstract

This paper studies a model learning and online planning approach towards building flexible and general robots. Specifically, we investigate how to exploit the locality and sparsity structures in the underlying environmental transition model to improve model generalization, data-efficiency, and runtime-efficiency. We present a new domain definition language, named PDSketch. It allows users to flexibly define high-level structures in the transition models, such as object and feature dependencies, in a way similar to how programmers use TensorFlow or PyTorch to specify kernel sizes and hidden dimensions of a convolutional neural network. The details of the transition model will be filled in by trainable neural networks. Based on the defined structures and learned parameters, PDSketch automatically generates domain-independent planning heuristics without additional training. The derived heuristics accelerate the performance-time planning for novel goals.