Large-Scale Environmental Mapping In Real-Time By A Robotic System

Number: 20190068940
Published: 2019-02-28
Filed: 2017-08-31
Assignee: DISNEY ENTERPRISES INC.
Inventors: Beardsley; Paul, Vempati; Anurag Sai, Nieto; Juan, Gilitschenski; Igor
CPC: H04N13/122; H04N13/271; G01C15/002; G01C21/20; G05D1/00; G05D1/0011; G05D1/0202; G05D1/0206; G05D1/0248; G05D1/027; G05D1/0274; G06T15/08; G06T17/00; G06T17/05; H04N13/25; H04N13/257; H04N13/296
Verdict: High Hardware
Source: Google Patents · FreePatentsOnline

The keeper's note

Large-scale environmental mapping real-time robotic system PGPUB dup (Beardsley).

Abstract

According to one implementation, a robotic system for performing large-scale environmental mapping in real-time includes a mobile reconnaissance unit having a color sensor, a depth sensor, and a graphics processing unit (GPU) with a GPU memory, and a navigation unit communicatively coupled to the mobile reconnaissance unit and having a central processing unit (CPU) with a CPU memory. The robotic system begins a three-dimensional (3D) scan of an environment of the mobile reconnaissance unit using the color sensor and the depth sensor, and generates mapping data for populating a volumetric representation of the environment. The robotic system continues the 3D scan of the environment using the color sensor and the depth sensor, updates the mapping data, and partitions the volumetric representation between the GPU memory and the CPU memory based on a memory allocation criteria.

Background

BACKGROUND

The generation of detailed and accurate representations of a local environment, using dense simultaneous localization and mapping (dense SLAM) methods, for example, can be important for robotics applications such as navigation and scene interpretation. Although one obstacle to successfully producing such environmental representations has been the processing overhead required by dense SLAM, advances in computing technology have made that particular obstacle less formidable. For example, powerful graphics processing units (GPUs) enabling dense SLAM algorithms to harness the power of parallelization are now widely available.

Several further challenges need to be addressed in order to make dense SLAM suitable for real-world applications, however. For example, conventional dense SLAM systems typically do not scale to large-scale environments because they are constrained by GPU memory, thus limiting the size of the area that can be mapped. Another limitation is the inability to handle rapid or abrupt camera motion, which is particularly problematic for agile aerial robotic vehicles, such as aerial drones, for example.SUMMARY

There are provided robotic systems and methods for performing large-scale environmental mapping in real-time, substantially as shown in and/or described in connection with at least one of the figures, and as set forth more completely in the claims.

Claims

1. A robotic system comprising: a mobile reconnaissance unit having a color sensor, a depth sensor, and a graphics processing unit (GPU) with a GPU memory; a navigation unit communicatively coupled to the mobile reconnaissance unit, the navigation unit having a central processing unit (CPU) with a CPU memory; the robotic system configured to: begin a three-dimensional (3D) scan of an environment of the mobile reconnaissance unit using the color sensor and the depth sensor; generate a mapping data for populating a volumetric representation of the environment produced using the GPU; continue the 3D scan of the environment using the color sensor and the depth sensor; update the mapping data based on the continued 3D scan; and partition the volumetric representation between the GPU memory and the CPU memory based on a memory allocation criteria; wherein the volumetric representation of the environment is stored partially in the GPU memory of the mobile reconnaissance unit and partially in the CPU memory of the navigation unit. 11. A method for use by a robotic system including a mobile reconnaissance unit having a color sensor, a depth sensor, and a graphics processing unit (GPU) with a GPU memory, and further including a navigation unit communicatively coupled to the mobile reconnaissance unit and having a central processing unit (CPU) with a CPU memory the method comprising: beginning a three-dimensional (3D) scan of an environment of the mobile reconnaissance unit using the color sensor and the depth sensor; generating a mapping data for populating a volumetric representation of the environment produced using the GPU; continuing the 3D scan of the environment using the color sensor and the depth sensor; updating the mapping data based on the continued 3D scan; and partitioning the volumetric representation between the GPU memory and the CPU memory based on a memory allocation criteria; wherein the volumetric representation of the environment is stored partially in the GPU memory of the mobile reconnaissance unit and partially in the CPU memory of the navigation unit.