Cloud-Based Collaborative 3D Mapping in Real-Time With Low-Cost Robots

This paper presents an architecture, protocol, and parallel algorithms for collaborative 3D mapping in the cloud with low-cost robots. The robots run a dense visual odometry algorithm on a smartphone-class processor. Key-frames from the visual odometry are sent to the cloud for parallel optimization and merging with maps produced by other robots. After optimization the cloud pushes the updated poses of the local key-frames back to the robots. All processes are managed by Rapyuta, a cloud robotics framework that runs in a commercial data center. This paper includes qualitative visualization of collaboratively built maps, as well as quantitative evaluation of localization accuracy, bandwidth usage, processing speeds, and map storage. Note to Practitioners-This paper presents an architecture for cloud-based collaborative 3D mapping with low-cost robots. The low-cost robots used in this work consist mainly of a mobile base, a smart phone class processor, an RGB-D sensor, and a wireless interface. Each robot runs its own visual odometry algorithm, which estimates the pose of the robot using the color and the depth frames (images) from the RGB-D sensor. The dense visual odometry algorithm presented herein uses no image features and requires no specialized hardware. In addition to pose estimation, the visual odometry algorithm also produces key-frames, which is a subset of frames that in a way summarizes the motion of the robot. These key-frames are sent to the cloud for further optimization and merging with the key-frames produced by other robots. By sending only the key-frames (instead of all the frames produced by the sensor), bandwidth requirements are significantly reduced. Each robot is connected to the cloud infrastructure using a WebSocket-based bidirectional full duplex communication channel. The cloud infrastructure is provided using Rapyuta, a Platform-as-a-Service framework for building scalable cloud robotics applications. The key-frame pose optimization and the merging processes are parallelized in order to make them scalable. The updated key-frame poses are eventually sent back to the robot to improve its localization accuracy. In addition to describing the architecture and the design choices, the paper provides qualitative and quantitative evaluations of the integrated system.