Whether it's automotive manufacturers or drone makers, the world is well on its way to a day when autonomous cars and drones will take on increasing roles, transporting commuters to work and playing major roles in logistics. This is raising new questions about the control of such autonomous technology.
The control of a Swarm of autonomous devices and vehicles is a critical question that has yet to be answered. One approach is to have a distributed approach, where every car and every drone is individually aware of its surroundings, individually directed where to go, and given the computational and AI capabilities to make it through the world efficiently and without accident.
Of course, these vehicles will incorporate a degree of peer-to-peer control, but primarily each machine will be doing its own calculations.
Another approach, and one advocated by a collection of faculty at Stanford University, is for swarms to be managed centrally, using applications running in large data centres. To explore this possibility, they have formed a new laboratory called Platform Lab to develop the infrastructure needed for these applications.
According to John Ousterhout, centralised control of self-driving cars and drones will allow an increased degree of coordination, with the potential to change how society functions on a daily basis.
One other advantage is the ease of creating applications. In a distributed control model, each device has limited information about the state of the world, and writing applications for each device can be difficult.
With a centralised approach, data from all the devices is collected in one place. With a big-picture view of the world, it is easier to control higher-level tasks like system-wide situational perception, decision-making and large-scale traffic planning.
By concentrating control into data centres, it is also possible to bring many more resources into play, including computing horsepower and large back-end datasets. This allows the implementation of more sophisticated collaborative behaviours for the entire swarm. It is also possible to leverage these resources to activate powerful machine learning algorithms, so the control system can learn and improve its behaviour.
The centralised approach also overcomes the ever-accelerating need for upgrades. If computational capabilities are built into the devices themselves, as systems and algorithms become more sophisticated, the computational hardware would need to be upgraded for each device, which can quickly become an expensive exercise.
In contrast, with centralised control, the vehicle is merely a tool, fitted with equipment that allows it to see the road and the skies, detect obstacles and other vehicles, and provide geolocation data. All this data is gathered and transferred back to the cloud, processed by powerful processors, and instructions then provided back to the individual vehicles.
“From a technology standpoint, it is attractive and easiest to centralize control – to amass data, plan and then disseminate a singular view to all devices,” says Ousterhout.
Devices, however, will retain local control of things like device stability and near-term collision avoidance. Such control needs microsecond or sub-millisecond response time and must happen on the device.
One major challenge for centralisation, however, is that much of the infrastructure does not yet exist, and the capabilities needed for swarm control is wide-ranging, including, but not exclusive to GPS, mapping, wireless communications, situational awareness and traffic coordination.
Another challenge is to provide a massive amount of computing with extremely low and predictable latency. That means leveraging new machine learning and artificial intelligence techniques to ensure planning and control happen fast without data inference.
The mission for the Stanford University team is to imagine how this centralised future might function and to determine what pieces exist, which need to be improved and what others are yet to be created to make it all function seamlessly. To this end, they are laying out the roadmap of the platform architecture, from the applications that will be needed to track packages, manage commutes, coordinate disaster relief and provide mapping to deep learning, adaptive scheduling and data-gathering tools.
The lab will also foresee the need for new technology like new hardware accelerators, better ways for computers to manage the many computing threads occurring simultaneously, rapid data storage and retrieval, and improved cluster scheduling necessary to execute the massive number of computations centralized control will demand.