As more sensors and communications features are added to automobiles, it’s clear that vehicles are already headed toward some level of autonomy, despite very valid concerns. Overcoming those concerns requires enormous processing horsepower and cloud communications, and Intel® GO™ provides the necessary hardware and software to accomplish this.
In last week’s update on Ilika’s thin-film batteries for IoT applications, it was mentioned in passing that vehicles are becoming just another IoT dot. That was an understatement: they’re actually already a really big, mobile and critically important IoT dot. With sensors under the hood feeding engine control units (ECUs), in-cabin sensors helping to make passengers comfortable and external sensors keeping the entire system and its cargo safe, it’s a small feat to imagine automobiles soon driving themselves, but we’re not there yet. Not by a long shot.
Last September, the U.S. Department of Transportation (DoT) adopted the Society of Automotive Engineers (SAE) International’s six levels of automation, and right now, we’re only at Level 2, pushing at Level 3 (Figure 1).
Figure 1. The US DoT adopted SAE International’s J3016 definition of six levels of vehicle autonomy. Right now the industry is at Level 2, pushing Level 3. (Image source: SAE International).
At Level 0, there is no assistance, only warnings. Level 1 includes features like adaptive cruise control, parking assistance and lane-keeping assistance. Level 2 enters the hands-off phase, but the driver must be ready to retake control of the vehicle should it not respond properly. At Level 3, the driver can take their eyes off the road, but should be alert enough to respond to a request to intervene. Level 4 allows the driver to sleep, while Level 5 is full automation.
While the industry is working toward Level 3, the hard part is getting from Level 3 to Level 4. Level 5 is exponentially more difficult. Just getting comfortable with Level 3 will require more research, more sensing, more processing and more communications between vehicles and infrastructure.
According to Intel, the amount of data an autonomous vehicle will generate will reach 4,000 Gbytes per day. To help deal with the processing and communication of that amount of data, and to reduce the barrier to entry for solution developers, the company announced the Intel® GO™ Autonomous Driving Solutions earlier this year (Figure 2).
Figure 2. Intel® GO™ Autonomous Driving Solutions enable the vehicular IoT dot such that it can sense and communicate as needed to reach higher levels of autonomy. (Image source: Intel Corp.)
Recognizing that the processing of data from LiDAR, radar, high-speed cameras and internal sensors will require a combination of high-performance parallel and sequential computing, the company turned to its own CPUs and field-programmable gate arrays (FPGAs). Specifically, Intel GO uses a combination of Intel® Atom® and Intel® Xeon® processors for automotive, with Intel® Arria® 10 FPGAs (Figure 3).
Figure 3. The Intel® GO™ platform combines Intel® Atom® or Intel® Xeon® processors for automotive, with an Intel® Arria® 10 FPGA to optimize sensor data processing. Shown is the Atom processor version. (Image source: Intel Corp.)
The logic behind the approach is that repetitive functions can be assigned to the FPGAs, while the CPU can handle the decision-making. Getting the mix right is critical, so Intel also provided a software development kit (SDK) to allow developers to delegate functions appropriately and easily.
The kit includes computer vision, deep learning and OpenCL tool kits to more quickly develop the middleware and algorithms. A sensor data-labeling tool allows for the creation of “ground truth” for deep-learning training and environment modeling. The SDK also has driving-targeted libraries, compilers, performance and power analyzers, and of course, debuggers. The tools emphasize a functional safety workflow, a “must have” if autonomous vehicles are ever to make it to the road, at any level.
For communication, the platform is looking to incorporate 5G, which appears to be the ultimate connector of all IoT dots. Between now and 5G’s availability, there is much research and data gathering to be done, and vehicle-compatible IoT solutions for that need to be made available for academia, small startups and large manufacturers.