Skip to main content


Scale AI into Next-Gen Robot Surgeons

AI surgical robots, robotic systems

The healthcare industry has always been on the cutting edge of AI. Take, for example, the work on an inferencing engine in 1972 at Stanford University. Known as the MYCIN expert system, it successfully leveraged AI to diagnose blood infections in patients based on reported symptoms and medical test results.

Now healthcare AI is at it again. In this case, it’s making surgical robots smart enough to perform automated tasks, as demonstrated by a research partnership between UC Berkeley, Intel® AI Labs, and Google Brain. One project is Motion2Vec, a semi-supervised representation learning algorithm deployed on a da Vinci surgical robot, learning how to suture wounds completely autonomously.

Motion2Vec is still being refined, but someday it could support AI surgical robots that are more precise than the most practiced human surgeons.

#Healthcare #AI is at it again. In this case, it’s making surgical #robots smart enough to perform automated tasks. @congatecAG via @insightdottech

The Hidden Costs of Robotic Systems

What’s limiting the advancement of fully autonomous surgical robots isn’t so much AI models but the complexity of underlying hardware, regulatory constraints, and the costs of each.

The control modules in these systems are often proprietary, relying on discrete, fixed-function CPUs, GPUs, crypto processors, or ASICs for each required task. The result is a complex system architecture comprising multiple components that not only limit what a robot can do today but also inhibit its ability to scale into the future.

To make matters worse, once these designs are certified to functional safety standards, they are almost set in stone. The time and cost of recertifying to medical safety standards trumps any advantage of upgradability. At that point, it often makes sense to build or buy a completely new system.

But with such high price tags, these systems should be upgradable—a truth that will become even more pointed as AI technology continues to evolve. The problem is that an AI surgical robot has so much going on it’s not really one system at all.

“On an AI robot system, there’s vision required to recognize what’s happening,” says Christian Eder, Director of Marketing at congatec AG, a leading supplier of embedded computer modules. “There is all this motion control and real-time processing necessary, and of course you have to maintain high levels of safety. So that’s an essential application: to combine things like vision and motion and safety.”

This is where 11th Gen Intel® Core vPro® and Intel® Xeon® W-11000E Series processors (previously known as Tiger Lake H) come in.

Integration: The Prognosis for Next-Gen Robotics

These 11th Gen processors overcome the challenges of multidisciplinary systems by essentially consolidating them in a 10 nm chipset. This includes heterogeneous multicore processing performance, low-latency and deterministic communications, and even functional safety measures, all in a configurable thermal design power (TDP) as low as 25 W.

On the processing front, devices in the portfolio offer up to eight CPUs. But it’s really the surrounding compute features that make 11th Gen processors stand out in complex use cases like surgical robots.

  • Integrated Intel® UHD Graphics support CV and/or AI workloads with a parallel option that executes these tasks more efficiently. The graphics units also free up CPUs for other tasks like control, network management, and general computing.
  • Hardware-accelerated partitions allow various cores, graphics units, and other components to operate as independent virtual platforms with Intel® Virtualization Technology.
  • Deep-learning inferencing performance achieves higher-performance instructions than previous generations via Intel® Deep Learning Boost.

But sheer performance isn’t the only way the new chipsets enable next-gen robotic systems. For example, once an output is generated by a Motion2Vec inference, that command must be executed on robotic motors and actuators in real time to guarantee procedures are performed in the correct order.

The 11th Gen Intel Series processors ensure that this hard real-time performance leveraging Time Sensitive Networking (TSN) and Intel® Time-Coordinated Computing (Intel® TCC). TCC synchronizes IP blocks within the processor and is supported by tools that reduce jitter and protect real-time applications from interference.

Other aspects of the new devices focus specifically on functional safety. Built-in hooks map processor hardware and firmware with the Intel® Functional Safety Essential Design Package (Intel® FSEDP), which significantly reduces safety certification effort.

The high level of hardware integration and on-ramp to FuSa compliance lends itself to simpler, more streamlined development—easing upfront component expenses and certification costs later. But what about the significant cost savings that can be achieved from upgrading, rather than reengineering, a robotic system?

Standard Modules Equals Development Savings

No amount of chipset performance or integration will protect from the cost of upgrading to a more functional robot. But new embedded hardware standards can.

The PICMG COM Express and COM-HPC specifications are embedded computer-on-module (COM) standards that leverage modular, two-board architectures. In both instances, the bottom card serves as an I/O pathway into a system like a robot—which allows the top processor module to be exchanged for one with more performance if the interfaces between the two remain compatible.

The only difference between the two standards is that COM Express serves existing designs while COM-HPC was architected to support next-generation interfaces, processors, and the higher TDPs that come with them.

As a result, developers who design around COM Express can easily adopt next-gen modules like the conga-TS570. Those starting anew can take advantage of 11th Gen Intel Core processors via solutions like the congatec conga-HPC/cTLH. And in the future, they can replace it with a COM-HPC module based on the next generation of chipsets without having to redesign the entire system.

“The beauty of these processors is that they operate within the same power envelope as previous-generation Intel Core processors, but we get much more performance. And we have this AI acceleration already on the processor,” Eder explains. “In just one low-power package that does not require any extra power supplies or any extra cooling, we can continue with the level we were at before, but with much more functionality.”

With an enabling hardware foundation in place, AI technologies like Motion2Vec should progress to a point that not only assists physicians but replaces them in some cases. This will make healthcare more accessible to all, both economically and geographically.

All these doc bots need is a more flexible brain.

About the Author

Brandon is a long-time contributor to going back to its days as Embedded Innovator, with more than a decade of high-tech journalism and media experience in previous roles as Editor-in-Chief of electronics engineering publication Embedded Computing Design, co-host of the Embedded Insiders podcast, and co-chair of live and virtual events such as Industrial IoT University at Sensors Expo and the IoT Device Security Conference. Brandon currently serves as marketing officer for electronic hardware standards organization, PICMG, where he helps evangelize the use of open standards-based technology. Brandon’s coverage focuses on artificial intelligence and machine learning, the Internet of Things, cybersecurity, embedded processors, edge computing, prototyping kits, and safety-critical systems, but extends to any topic of interest to the electronic design community. Drop him a line at, DM him on Twitter @techielew, or connect with him on LinkedIn.

Profile Photo of Brandon Lewis