Autonomous robots have been working in collaboration with humans in the field of industrial manufacturing for more than half a century. Since the world's first industrial robot was developed and put into use in the 1950s, enterprises have entrusted robots with cumbersome and dangerous tasks, allowing workers to focus on more specialized work. Nowadays, the application of advanced robot technology is no longer confined to the industrial field but has also expanded to numerous vertical industries such as healthcare, retail, and agriculture.
Meanwhile, technological breakthroughs in fields such as artificial intelligence and machine learning have given rise to a new generation of more intelligent robots. They are no longer confined to performing repetitive tasks but can undertake more complex work. For instance, with the aid of technologies such as computer vision and autonomous movement, robots can handle various tasks including product assembly, quality inspection, advanced threat identification and response, etc.
In short, intelligent robots have become a core asset for strengthening the modern labor force. Their high-precision features and nearly unlimited potential for productivity improvement are irreplaceable. However, as enterprises' demands for robot assistants continue to upgrade, the design difficulty of such systems has grown exponentially, and there is an urgent need for low-latency and high-performance hardware such as Field Programmable Gate arrays (FPgas) to provide technical support.
The increasingly prominent design challenges
Intelligent robots equipped with artificial intelligence need to be fitted with more sensors and actuators compared to traditional robots, including cameras, lidars, radars, inertial measurement units (IMUs), motor encoders, pressure sensors and other components. Meanwhile, the robot also needs to complete more complex computing tasks in real time, such as 3D vision processing, simultaneous localization and mapping (SLAM), and grasping point calculation.
This requires that the hardware of the relevant systems not only have more input/output interfaces (I/O) to adapt to various sensors, but also be equipped with more powerful processing modules (such as central processing units (CPU), graphics processing units (GPU), and neural network processing units (NPU)) to achieve more complex computing functions. However, the challenge faced by designers is that relying solely on processing modules like cpus, it is difficult to connect with various sensors required by the robot system, nor can it efficiently handle the massive amount of raw data collected by the sensors.
The reasons for this are, on the one hand, that the number of I/O interfaces and the degree of specialization of cpus often fail to meet the demands of developers. Moreover, simply adding interfaces to the processor incurs high costs - physical interfaces need to maintain a specific size to achieve functions, and adding new interfaces means occupying more chip area. This is fundamentally different from the logic units that can be easily miniaturized and expanded in advanced manufacturing processes.
Even if the CPU can provide sufficient adaptive I/O interfaces to connect with intelligent robots and directly transmit the large amount of raw data collected by sensors to the processing unit, there is still a problem of low energy efficiency. Furthermore, the CPU is not designed for the real-time processing tasks required by intelligent robots. If core tasks such as sensor fusion are handled by the CPU, it will cause significant delays in the system and greatly reduce the operational efficiency of the robot.
Fortunately, hardware designers and developers are concentrating on developing various innovative products to make up for the above-mentioned technical shortcomings, and FPGA is one of them.
FPGA: A highly valuable hardware solution
FPGA is a highly flexible semiconductor device that can serve as a "bridge" between sensors, actuators and cpus, providing developers with various and numerous I/O interfaces required for the connection of intelligent robot systems. Meanwhile, with the real-time computing power near the sensor end, FPGA can undertake the underlying dedicated processing tasks of various sensors, release system computing resources, and help create more intelligent and responsive robots that enterprises need.
After the first layer of data processing is completed by the FPGA, the data will be transmitted to the CPU through standardized high-bandwidth channels. Through this task splitting method, FPGA can share part of the computing load with CPU, save energy consumption to support higher-order computing tasks such as trajectory planning, cluster analysis, and object detection, allowing CPU to focus on handling optimization and decision-making tasks that are difficult to achieve at the hardware level.
This hardware architecture can also help developers overcome the following types of technical challenges:
Connectivity: FPGA hardware has an extremely high degree of customization and can provide more I/O interfaces than cpus. Developers can connect and control more sensors and actuators through various interfaces such as Ethernet, Serial Peripheral Interface (SPI), High Definition Multimedia Interface (HDMI), and Mobile Industry Processor Interface (MIPI), and the cost is much lower than adding new interfaces to the main processing unit. In addition, FPgas also support multiple voltage levels and non-standard communication protocols, providing developers with more options to adapt to different application scenarios.
Energy consumption: FPGA can achieve hardware-level parallel computing near the robot sensors. By processing the data locally in real time and then transmitting it to the CPU, it effectively reduces the overall energy consumption of the system.
Latency: The high-speed computing power of FPGA can accelerate the processing of core tasks such as sensor fusion - this task can integrate the collected data from different sensors such as cameras and lidars to form a complete environmental perception picture, thereby enhancing the judgment accuracy and decision-making ability of robots. Take the computing speed as an example. The VLP16 liDAR sensor transmits 384 sets of distance data to the network every 1.32 milliseconds, while the FPGA only needs about 0.32 milliseconds to complete the processing of this batch of data, with a computing speed of 100 million times per second.
Relying on the various technical advantages of FPGA, designers can flexibly install various sensors according to their needs, break through the performance upper limit of intelligent robots, and effectively solve the energy consumption and delay problems of the system at the same time.
Join hands to create smarter robots
As the demand for smarter and faster robots in various industries continues to rise, developers are facing new challenges: designing robot systems with better performance without depleting resources. To achieve this goal, developers are increasingly relying on hardware design and manufacturing enterprises to continuously upgrade the core components of robots. Both sides are working towards the common goal of "enhancing robot performance while reducing costs, energy consumption and latency", making the future development of the robotics field full of infinite possibilities.
