Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

coVoxSLAM: GPU Accelerated Globally Consistent Dense SLAM

Emiliano Höss, Pablo De Cristóforis

Abstract

A dense SLAM system is essential for mobile robots, as it provides localization and allows navigation, path planning, obstacle avoidance, and decision-making in unstructured environments. Due to increasing computational demands the use of GPUs in dense SLAM is expanding. In this work, we present coVoxSLAM, a novel GPU-accelerated volumetric SLAM system that takes full advantage of the parallel processing power of the GPU to build globally consistent maps even in large-scale environments. It was deployed on different platforms (discrete and embedded GPU) and compared with the state of the art. The results obtained using public datasets show that coVoxSLAM delivers a significant performance improvement considering execution times while maintaining accurate localization. The presented system is available as open-source on GitHub https://github.com/lrse-uba/coVoxSLAM.

coVoxSLAM: GPU Accelerated Globally Consistent Dense SLAM

Abstract

A dense SLAM system is essential for mobile robots, as it provides localization and allows navigation, path planning, obstacle avoidance, and decision-making in unstructured environments. Due to increasing computational demands the use of GPUs in dense SLAM is expanding. In this work, we present coVoxSLAM, a novel GPU-accelerated volumetric SLAM system that takes full advantage of the parallel processing power of the GPU to build globally consistent maps even in large-scale environments. It was deployed on different platforms (discrete and embedded GPU) and compared with the state of the art. The results obtained using public datasets show that coVoxSLAM delivers a significant performance improvement considering execution times while maintaining accurate localization. The presented system is available as open-source on GitHub https://github.com/lrse-uba/coVoxSLAM.

Paper Structure

This paper contains 17 sections, 2 equations, 7 figures, 1 table.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. SYSTEM ARCHITECTURE DESCRIPTION
  4. Data Layout
  5. Data Structures
  6. TSDF Representation
  7. Point cloud Integration
  8. ESDF computation
  9. Backend
  10. EXPERIMENTS AND RESULTS
  11. TSDF Timings
  12. ESDF Timings
  13. Backend Timings
  14. Hash table stress testing
  15. RMSE
  16. ...and 2 more sections

Figures (7)

  • Figure 1: A reconstruction resulting from a 400 m-long MAV flight through the search and rescue training site discussed in Sec. \ref{['results']}. In the foreground a large pile of rubble from a collapsed structure is visible. The trajectory (green) also contains indoor-outdoor transitions through a building.
  • Figure 2: Data flow pipeline of the system, using two queues for the front end and the back end to improve reliability. The mapper reads from the queue when a new pointcloud is available. When the submap threshold is reached, it enqueues the current submap to be finalized by the ESDF Integrator, ends the backend and returns a new empty submap to the TSDF Integrator. Otherwise, it returns the current submap. The size of both queues can be modified as a parameter of the system.
  • Figure 3: The estimated trajectories of Flight 1 dataset by Voxgraph (blue), coVoxSLAM on PC (orange), coVoxSLAM on Jetson Xavier AGX (green) and RTK-GNSS measurements (red) used as ground truth to evaluate the system.
  • Figure 4: Execution times for the TSDF Integrator, including TSDF + Color.
  • Figure 5: Speep-up of execution times of ESDF integration in four different datasets.
  • ...and 2 more figures