The Berkeley artificial intelligence research laboratory is working on the fogros project. The project is part of the latest version of the open source robot operating system ROS 2 humble hawksbill and the eighth version of ROS 2 The progress of server-side computing enables things such as cloud games to be realized with minimal delay, which can also be applied to robot operation
Due to the weight and power requirements, the onboard computing power of robots is often limited, and they rarely have hardware accelerators, such as GPU, TPUs or FPGA. But many robotic algorithms and recent advances have benefited from high-end computers and hardware accelerators. Berkeley artificial intelligence research laboratory envisages using cloud computing to accelerate slow computing, which can enable robots to do more things in the same time.
In March this year, the team quietly provided a preview version of fogros 2 through GitHub. Today, it is open to everyone, with many performance improvements based on cloud computing. Play [Xbox] with on your Smartphone( https://microsoft.pvxt.net/e4yLO ) Like the game, the basic principle here is to provide a method to perform complex tasks on the robot without the same complex airborne processing. If you can complete the task through the remote server, you can save size, weight and cost.
In a recently published paper, the team proved that the performance obtained by using cloud computer can overcome network delay and greatly accelerate the performance of robot. In the example, fogros 2 reduces the slam delay by 50%, reduces the capture planning time from 14 seconds to 1.2 seconds, and improves the motion planning speed by 28 times. Compared with the alternative, fogros 2 reduces the network utilization by 3.8 times.
Goldberg pointed out that such a platform could open up more possibilities for robotics. It may be beneficial to other computing intensive tasks of robots, such as random programming, and promote supervised and unsupervised deep learning of multiple robot tasks. Future versions of the plan can be opened to more platforms, including Google cloud and azure.