Spatio-temporal voxel layer: A view on robot perception for the dynamic world

Abstract

The spatio-temporal voxel grid is an actively maintained open-source project providing an improved three-dimensional environmental representation that has been garnering increased adoption in large, dynamic, and complex environments. We provide a voxel grid and the Costmap 2-D layer plug-in, Spatio-Temporal Voxel Layer, powered by a real-time sparse occupancy grid with constant time access to voxels which does not scale with the environment’s size. We replace ray-casting with a new clearing technique we dub frustum acceleration that does not assume a static environment and in practice, represents moving environments better. Our method operates at nearly 400% less CPU load on average while processing 9 QVGA resolution depth cameras as compared to the voxel layer. This technique also supports sensors such as three-dimensional laser scanners, radars, and additional modern sensors that were previously unsupported in the available ROS Navigation framework that has become staples in the roboticists’ toolbox. These sensors are becoming more widely used in robotics as sensor prices are driven down and mobile compute capabilities improve. The Spatio-Temporal Voxel Layer was developed in the open with community feedback over its development life cycle and continues to have additional features and capabilities added by the community. As of February 2019, the Spatio-Temporal Voxel Layer is being used on over 600 robots worldwide in warehouses, factories, hospitals, hotels, stores, and libraries. The open-source software can be viewed and installed on its GitHub page at https://github.com/SteveMacenski/spatio_temporal_voxel_layer .

VDBFusion: Flexible and Efficient TSDF Integration of Range Sensor Data

“…To implement our system, we build upon the VDB data structure implemented in the OpenVDB [ 15 ] library. In line with our work, Macenski et al [ 48 ] developed a spatio-temporal voxel system for 3D mapping making use of OpenVDB [ 15 ]. This system encodes sensor observations into an occupancy grid map that additionally implements voxel decay and decay acceleration [ 48 ].…”

“…In line with our work, Macenski et al [ 48 ] developed a spatio-temporal voxel system for 3D mapping making use of OpenVDB [ 15 ]. This system encodes sensor observations into an occupancy grid map that additionally implements voxel decay and decay acceleration [ 48 ]. More recently, Besselmann et al [ 49 ] have also shown promising using the VDB data structure to implement an octree-inspired representation similar to Octomap [ 6 ].…”

“…To the best of our knowledge, OpenVDB is the only well-supported, open source library dedicated to volumetric data applications. Although OpenVDB [ 15 ] has been open source for almost a decade, the robotics community paid little attention to it [ 48 , 49 ] in relation to the benefits that the library provides. Potentially, this is rooted in the lack of common vocabulary between communities.…”

Consideraciones de implementación para algoritmos de exploración robótica en plataformas de procesamiento limitado

“…Actualmente, se está trabajando en la implementación en CUDA (Compute Unified Device Architecture en inglés), del algoritmo recientemente presentado: STVL (Macenski, Tsai, & Feinberg, 2020). Se espera correr el mismo a bordo de una Computadora Jetson Nano, la cual integra una GPU embebida.…”

SLAM Toolbox: SLAM for the dynamic world

“…Retail and warehouse spaces can change drastically throughout the year and the state of roadways can be changing by the hour. Much work has been made to address changing environments in robot perception (Macenski, Tsai, et al, 2020), but less has been built in open-source to represent maps of dynamic spaces.…”