Autonomous robots have been working alongside humans in Industrial Manufacturing for more than half a century. Ever since the world's first Industrial robot was developed and deployed in the 1950s, businesses have been offloading tedious or dangerous tasks to these machines to free up their workers for more nuanced responsibilities. Today, the use of advanced robotics is no longer limited to the Industrial sector; it has expanded into many other verticals, including Healthcare, Retail and Agriculture.
Moreover, technological advancements—especially in the field of AI and machine learning—have spurred a new generation of even smarter robots that go beyond repetitive tasks to perform far more sophisticated functions. For example, through capabilities like computer vision and autonomous mobility, they can perform everything from product assembly to quality control to advanced threat detection and response.
In short, these smart robots have become an indispensable asset to bolster modern workforces, offering a high level of precision and a near-unlimited capacity for productivity. However, as businesses begin to demand more from their robotic assistants, these systems are getting exponentially harder to design. They need the support of low latency, high power hardware like Field Programmable Gate Arrays (FPGAs).
Growing design challenges
Smart, AI-powered robots require significantly more sensors and actuators than their predecessors, including components like cameras, LiDAR, radar, inertial measurement units (IMUs), motor encoders and pressure sensors. They also need to perform much more complex computing tasks—like 3D vision processing, Simultaneous Localization and Mapping (SLAM) and pick point calculations—in real time.
As a result, building these systems requires the use of hardware with more Inputs/Outputs (I/Os) to accommodate more sensors, as well as higher performance processing modules (e.g., CPUs, GPUs or NPUs) to perform more advanced computing functions. The challenge? It’s often unfeasible for designers to rely solely on processing modules like CPUs to interface with the multitudes of sensors required in these systems and process all of their raw data.
One logistical reason for this is that CPUs may not provide as many or as specialized I/O ports as developers require, and simply adding more ports to these processors can be expensive. This expense stems from the fact that physical connectors like I/Os have to remain a certain size to be functional, so adding ports means taking up more space on the chip. This differentiates them from logic blocks, which can be easily shrunk or scaled down in advanced processing nodes.
Even if a CPU does provide enough of the right I/Os for smart robot connectivity, transmitting high volumes of raw sensor data directly to these processing units is not energy efficient. Not only that, but CPUs are also not optimized for the real-time processing that smart robots require. For example, delegating critical tasks like sensor fusion to CPUs would introduce noticeable latency to these systems and significantly slow down their operations.
Thankfully, hardware designers and developers are focused on innovating products that help bridge these gaps—including FPGAs.
FPGAs: A valuable hardware solution
FPGAs are flexible semiconductors that act as a “bridge” between sensors, actuators and CPUs, enabling developers with the number and type of I/Os required to connect their smart robotics systems. Thanks to their real-time, near sensor computing capabilities, FPGAs can also take on low level, sensor-specific tasks to free up system capacity and support the smarter, more responsive robots that businesses need.
Once the data has gone through the first layer of FPGA-based processing, it’s then transferred to the CPU via a standard, high bandwidth channel. By partitioning the processing tasks of smart robots in this way, FPGAs take some of the load away from the CPU and save energy for higher-level computations like trajectory planning or clustering and object detection. This allows the CPU to focus on the types of optimization and decision tasks that are harder to perform at the hardware level.
This setup also helps developers overcome challenges related to:
- Connectivity: FPGA hardware is highly customizable and generally offers more I/Os than CPUs. This allows developers to connect and control more sensors and actuators using I/Os like Ethernet, SPI, HDMI, MIPI and more, at a lower cost than adding ports to the main processing unit. FPGAs also support various voltage levels and non-standard protocols, expanding developers’ options to meet the needs of different applications.
- Power: FPGAs offer hardware-based parallel processing nearer to robotic sensors. By performing real-time local computing to process data and send it to the CPU, they help reduce the energy consumption of systems.
- Latency: FPGAs’ processing speed accelerates critical tasks like sensor fusion, which merges data from different sensors like cameras and LiDAR to paint a complete picture, improving the accuracy and decision-making of robots. To showcase their processing speed, imagine that 384 pieces of distance data from a VLP16 lidar sensor are being sent over the network every 1.32 milliseconds. An FPGA only takes about 0.32 milliseconds to process this data, at a speed of 100 million operations per second.
Through FPGAs’ capabilities and benefits, designers have the freedom to incorporate as many varied sensors as they need to push the limits of intelligent robots, all while addressing power and latency constraints.
Collaborating on Smarter Robots As the demand for smarter, faster robots increases across industries, developers are being challenged to design higher performance systems without maxing out their resources. To achieve this, they are relying on hardware designers and manufacturers to improve the underlying components used to build these machines. With both parties working toward the end goal of optimizing robots’ capabilities while reducing costs, power and lag, the possibilities for the future of this field are limitless.
