Cloud-based Parallel Implementation of SLAM for Mobile Robots

Abstract

Simultaneous Localization and Mapping (SLAM) for mobile robots is a computationally expensive task. A robot capable of SLAM needs a powerful onboard computer, but this can limit the robot's mobility because of weight and power demands. We consider moving this task to a remote compute cloud, by proposing a general cloud-based architecture for real-time robotics computation, and then implementing a Rao-Blackwellized Particle Filtering-based SLAM algorithm in a multi-node cluster in the cloud. In our implementation, expensive computations are executed in parallel, yielding significant improvements in computation time. This allows the algorithm to increase the complexity and frequency of calculations, enhancing the accuracy of the resulting map while freeing the robot's onboard computer for other tasks. Our method for implementing particle filtering in the cloud is not specific to SLAM and can be applied to other computationally-intensive tasks.