Designing efficient autonomous navigation systems for mobile robots involves consideration of the robotís environment while arriving at a systems architecture that trades off multiple constraints. We have architected a navigation framework for socially-aware autonomous robot navigation, using only the on-board computing resources. Our goal is to foster the development of several important service robotics applications using this platform. Our framework allows a robot to autonomously navigate in indoor environments while accounting for people (i.e., estimating the path of all individuals in the environment), respecting each individualís private space.
In our design, we can leverage a wide number of sensors for navigation, including cameras, 2D and 3D scanners, and motion trackers. When designing our sensor system, we have considered that mobile robots have limited resources (i.e., power and computation) and that some sensors are costlier than others (e.g., cameras and 3D scanners stream data at high rates), requiring intensive computation to provide useful insight for real-time navigation. We tradeoff between accuracy, responsiveness, and power, and choose a Hokuyo UST-20LX 2D laser scanner for robot localization, obstacle detection and people tracking. We use an MPU-6050 for motion tracking.
Our navigation framework features a low-power sensor system (< 5W) tailored for improved battery life in robotic applications while providing sufficient accuracy. We have completed a prototype for a Human Support Robot using the available onboard computing devices, requiring less than 60W to run. We estimate we can obtain similar performance, while reducing power by ~60%, utilizing low-power high-performance accelerator hardware and parallelized software.
Buscar en Dimension