A Closer Look at Cloud Robotics – The Distant Future of Robotics

As early as 1994, when the Internet was still in the nascent state, there were examples of industrial robots connected to the Internet, mainly for rudimentary remote operation without proper cybersecurity measures. Since then, the advancement in cloud computing has brought huge enhancements to the robotics industry. Instead of relying completely on the computing capabilities of local servers or machines, robots can leverage a network of remote computing, storage, and data resources to achieve performance gains and efficiency that would not be possible otherwise. Robots can be manipulated by a centralized control center, learn from different application scenarios, and adjust according to demands in near real time. The reality on the ground, however, is murky. The term has been thrown around in a generic manner, as the industry at large is trying to identify the ideal implementation approach and architecture framework.

It is, therefore, not surprising that cloud robotics is brought up frequently by cloud or connectivity vendors, but less so by companies like ABB, KUKA, and other robotics hardware vendors. This does not mean that robotics hardware vendors are against cloud-connected robots. In fact, most robotics vendors have an in-house cloud platform, such as ABB Ability, KUKA Connect, Yaskawa Drive Cloud, and FANUC FIELD. However, these cloud platforms are often designed for on-site maintenance and troubleshooting, operate in silos and have no communication with similar systems in different geographical locations. The latter is a deliberate choice based on security concerns, but it means that these robotics systems are still limited to local storage, computing, and data resources.Beyond this, it prevents the robots from integrating with other connected assets, systems, and platforms.

Before any cloud robotics system can be implemented, there are several fundamentals that must be met:

• Constant Internet Connectivity: Internet connectivity can be either Wi-Fi for high throughput indoor communication, Ethernet for wired time-sensitive networking, or cellular wireless for outdoor deployment. The connectivity also needs to be secure and reliable, low in latency, and seamless during handover, which is challenging with LTE, but possible with 5G.
• Remote Computing, Storage, and Data Resources: Most, if not all, industrial and commercial robots are still controlled by controllers or software located in on-premise servers. This does not comply with the definition of cloud computing, where robots can leverage a network of remote computing, storage, and data resources. Data resources should go beyond those collected within the four walls of the premises and ideally be gathered from all business units. This will enable robots to learn from different situations and be able to make better decisions.
• Data-Driven Robotics Intelligence: If the robot can already maximize its capabilities without relying on any data, then there is no real advantage to hooking up the robot to a cloud platform. Hence, one key purpose of the cloud robotics solution is to deliver improvements based on the training and inference of artificial intelligence models hosted in the cloud. These capabilities can either be related to knowledge representation, machine vision, natural language processing, or autonomous localization and navigation.
• Strong Cybersecurity Capability: A cloud robotics platform enables the transmission of mission- and business-critical information and the control and manipulation of robots over the air. The lack of a strong cybersecurity protocol can easily lead to hacking or abusing vulnerabilities of driverless vehicles, drones, or any autonomous systems, which can be dangerous and even fatal. This has discussed in a previous executive foresight.
• Support for a Fleet of Robots: Technically, a robot arm connected to the cloud can also achieve performance gains and efficiency, especially if networked with collaborative robots that work in conjunction with human workers, as well as in machine vision for inspection, object recognition, and monitoring, but a cloud robotics platform must be able to support large numbers of robots to bring economies of scale. Unlike a virtual machine, the container architecture of cloud computing brings extra flexibility and scalability, allowing the workload to be shared easily between different data centers.

Once the aforementioned key pillars are established, these are the main implementation approaches, segmented based on level of customization for the robotics hardware, and the locations of use cases:

The glaring obstacle to the cloud robotics business model is the level of customization versus generalization. Most robotics use cases may use the same generic hardware, such as SCARA, 6-axis arm, or delta robot, but they have highly specialized on-site requirements or Key Performance Indicators (KPIs). The rest may simply require fully customized hardware at the start. Akin to all cloud solutions, economies of scale are critical. The investment of remote computing and storage resources will only start to make economic sense once a large fleet of robots is using those resources. Since most robotics use cases are highly specialized, traditional robotics vendors are less inclined to make that investment and would rather focus on case-by-case deployment scenarios.

As with all other industries in places where traditional vendors lag behind, software companies and startups pick up the slack. CloudMinds, a Santa Clara-based startup, raised US$3 million in seed funding with the goal of introducing a cloud-based secure AI platform for robotics. Since then, Neurala, Tend, and Vicarious Systems are some of the startups that have raised significant amounts of funding for creating cloud platforms that integrate images, videos, and data from passive and active sensors to provide intelligence to robots. Unfortunately, a lot of these efforts focus on the cloud platform and have yet to work closely with hardware vendors.

On the other hand, SAP, Rockwell Automation, and Siemens are partnering with many established robotics vendors, such as FANUC, KUKA, and Fetch Robotics, to standardize and deploy cloud platforms that incorporate everything within the factory, which includes all robots. ABI Research envisions that a revenue share model between the cloud platform vendors and the hardware suppliers will be the ideal path. Non-exclusive partnerships will ensure that end users who are existing clients of these cloud platforms can integrate all aforementioned robotics hardware as they see fit on a bundled pay-per-use model, and the revenue can be shared between hardware suppliers that offer Robotics-as-a-Service(RaaS), and cloud platform vendors. Partnerships with software companies with secure networking expertise and industrial know-how, such as Telit or PTC, will bring extra layers of secured connectivity.

While it is still too early to determine whether cloud robotics will be the dominant approach in the industry, there are some encouraging signs in the market that will continue to drive the development of cloud robotics in the right direction. Network slicing in 5G is one of the emerging technologies that will be built into the connectivity layer, fulfilling high-quality network and cybersecurity KPIs. The industry has certainly moved a long way since website-based robot teleoperating.