Massively Parallel Construction of the Cell Graph

TL;DR

This paper introduces parallel algorithms for constructing cell graphs in motion planning on GPUs and presents a new GPU-based dictionary data structure using search trees for efficient operations on long vectors.

Contribution

It provides the first parallel algorithms for cell graph construction on GPUs and introduces a novel GPU dictionary implementation using search trees.

Findings

Efficient GPU algorithms for cell graph construction.

A new GPU dictionary data structure with search trees.

Improved performance over traditional hash table methods.

Abstract

Motion planning is an important and well-studied field of robotics. A typical approach to finding a route is to construct a {\em cell graph} representing a scene and then to find a path in such a graph. In this paper we present and analyze parallel algorithms for constructing the cell graph on a SIMD-like GPU processor. Additionally, we present a new implementation of the dictionary data type on a GPU device. In the contrary to hash tables, which are common in GPU algorithms, it uses a search tree in which all values are kept in leaves. With such a structure we can effectively perform dictionary operations on a set of long vectors over a limited alphabet.