The AMD platform approach to robotics delivers significant productivity gains
Developments in robotics continue to transform industrial automation. For manufacturers, automation brings productivity gains measured in finished units.
Productivity is also important for the OEMs developing the automation equipment. OEMs are now shifting toward using common hardware and software platforms to develop new robotic solutions.
Standardizing on these common platforms accelerates new product development, providing OEMs with their own productivity boost.
OEMs will need to consider what features they need for their automation equipment, as a robot’s design will vary by application. For example, a welding robot differs significantly from a pick-and-place robot, but most robots will feature at least one end effector. Exchangeable end effectors conduct a specific task, while the bulk of the robot – often referred to as the manipulator – provides mobility, enabling the end effector to move into position.
However, there are also some similarities between robots. These similarities enable a modular approach to robot design based on common components. All robots rely on sensors and actuators. Common sensors include force sensors, position sensors and image sensors. Actuators fall into three broad categories based on how they convert energy into actions. These categories include electric, hydraulic and pneumatic.
The sensors and actuators interface to the robot through a central controller. Depending on the flexibility required, the controller can be custom hardware or an industrial PC. OEMs are keen to optimize controller design by maximizing flexibility while following a platform design philosophy.
Designing robot controllers
The controller provides several functions, including high-level processing, such as network communications and managing the user interface. The controller will also perform low-level tasks such as sensor interface and motor control.
Motors used in robotics range from AC, DC, stepper and servo. Each type of motor requires its own motor control algorithm. Fast, real-time responsiveness is an important requirement in robotics, which puts pressure on the underlying hardware.
The application-specific nature of robots means that the controller will often need to interface to different sensors and actuators and control different motors in diverse ways. This requires a flexible and modular approach to hardware and software design.
Each new generation of robots is more capable than the last. OEMs are integrating more sensors with more intelligence and more operational flexibility. Collaborative robots, also known as cobots, are good examples of where this trend is going. Cobots are trained to perform manual tasks and operate alongside people. They are more flexible than other robots but have restricted speed and payload due to safety requirements.
Robot operating systems
The robot operating system (ROS) addresses the need for modular software solutions in robotics. This popular open-source software framework includes libraries for the features required in modern robotics. These features include navigation, manipulation and sensor fusion.
The hardware that is able to run ROS is varied. General-purpose processors are one option and can give reliable performance. The ROS 2 version of the framework also supports microROS, designed to run on microcontrollers that host a real-time operating system (RTOS). This extends the scope of ROS to any application where robotics is making an impact.
This flexible and extensible framework is popular and enables many OEMs to get to market faster. However, like many software frameworks, delivering real benefits requires hardware that can match this flexibility and extensibility.
Robots are becoming increasingly software-defined, while still relying heavily on the underlying hardware. To make the hardware as flexible and modular as the software, it needs to offer more than general-purpose computing.
Programmable hardware for robotics
Programmable logic, such as field programmable gate arrays (FPGAs), has evolved over many years to become comprehensive hardware platforms. An FPGA can also meet the need for high performance and real-time responsiveness thanks to its parallel architecture.
The software tools for FPGAs differ from those used for general-purpose processors and microcontrollers. Modern FPGAs now integrate these elements and are more capable hardware platforms. This means the tools have also developed.
Modern system-on-modules (SOMs) like the AMD Kria™ K24 and K26 adaptive SOMs make good hardware platforms for modern robotic applications. These SOMs are based on the AMD Zynq™ UltraScale+™ MPSoC (multiprocessor system-on-chip).
The Kria module from AMD
Based on the AMD Zynq™ UltraScale+™ MPSoC, the Kria K24 and K26 adaptive system-on-modules with baseboard now combine with the Robot Operating System (ROS) to provide a solid yet flexible platform for robotics development. This powerful SoC features a dual-core Arm® Cortex®-R5F and four Arm Cortex-A53 cores. The SOM also includes DDR4 memory, eMMC memory, expansion connectors, a TPM 2.0 security module and a power solution. The SOM is designed to be used at multiple points in robotic applications.
The Kria module in detail
The Kria system-on-modules and baseboard provide high performance interfaces needed by robotic applications.
To support the robotics sector, AMD has developed the Kria robotics stack. This is a set of libraries and utilities built on ROS 2 and the ROS SDK. The Kria robotics stack is aimed directly at roboticists by removing and abstracting the low-level complexity required to use the SOM, leaving robotics engineers more time to focus on their applications.
By standardizing on the AMD Kria robotics stack and the AMD Kria K24 and K26 Adaptive SOMs, engineers can be more productive while also gaining access to hardware that will accelerate their robotics application. The AMD Kria robotics stack supports production-grade reference designs that engineers can port to the hardware, and is as simple as using an app from a mobile app store.
Faster robotics design with reference solutions
The Kria robotics stack application reference designs come pre-designed and pre-built, so roboticists do not need to do any low-level FPGA design to use them. Examples of the reference designs already available include TSN Communications, Perception and 10GigE vision camera.
AMD has developed the KR260 Robotics Starter Kit to further support development. This includes the software development environment based on Kria robotics stack and a carrier card for the K26 adaptive SOM.
The carrier card includes a high-performance vision sensor interface, the SLVS-EC Rx. This interface is optimized for image sensing in industrial applications. A further three USB 3.0 ports provide camera interfaces and there is also a DisplayPort 1.2a interface.
Real-time networking is supported by four RJ-45 Ethernet ports (10/100/1000). One SFP+ cage (10G) for 10GigE vision is also on the card. Expansion ports compatible with Pmod and Raspberry Pi are included.
AMD is actively supporting the ROS 2 Hardware Acceleration Working Group. The group is focused on reducing the time needed to compute Gazebo (simulation) and ROS 2 flows on accelerated hardware. AMD is contributing design tool integration, reference examples, testing environments and more.
Learn more by visiting the AMD Kria webpage.
AMD, and the AMD Arrow logo, Kria, UltraScale+, Zynq, and combinations thereof are trademarks of Advanced Micro Devices, Inc. Other product names used in this publication are for identification purposes only and may be trademarks of their respective owners.