Artificial Intelligence at the Edge
Artificial Intelligence (AI) & Machine Learning (ML) are shaping and advancing complex automated technology solutions. Having these capabilities integrated into hardware allows for solutions that can recognize objects in an image, analyze and detect abnormalities in patterns or find key phrases. These functionalities can be paramount for applications including, but not limited to, autonomous vehicles and robots, monitoring devices for IoT networks, and voice & language processing applications. Nearly any industry can take advantage of this powerful technology, including manufacturing, oil and gas, security and defense, agriculture and automotive.
A rugged & compact platform that can provide AI & ML in a single small package is available from Gateworks. Gateworks single board computers can be paired with a Google Coral Tensor Processing Unit (TPU).
Gateworks SBC & Google Coral Highlights
- Rugged SBC & TPU Solution
- Accelerates neural networks with edge TPU
- Performance: 4 TOPS (Trillion Operations Per Second)
- Half-size Mini-PCIe Form Factor (PCIe Gen2 x1)
- Supported Framework: TensorFlow Lite
- Works with AutoML Vision Edge
- Power: 2 Watts Typ, 4 Watts Max
- -20 to +70C Operating Temperature
Data can be analyzed in many places including the cloud or locally. While the cloud offers larger processing power, it requires large amounts of bandwidth which may not always be available. In these cases, a locally available hardware accelerated TPU can handle the task. The Google Coral TPU is able to run on a Gateworks SBC. The SBC can utilize many different sensor inputs, including a camera for object recognition. Results from the inferences and local analysis can then be communicated to the cloud or acted on immediately. Typical cloud communication channels include, but are not limited to: CATM1 Cellular, 4G LTE & 5G Cellular, WiFi, Iridium Satellite.
Local AI Hardware
Gateworks SBCs are the perfect platform for an artificial intelligence solution. The Gateworks SBC allows for both sensor and cloud communication via different technologies.
The heart of the AI solution is the Google Coral Tensor Processing Unit. This is available in a half size Mini-PCIe card which fits into nearly all Gateworks SBCs (adapter may be required).
Gateworks SBC’s feature Mini-PCIe expansion slots which allow support of a variety of wireless standards such as Cellular (LTE/CATM1/4G LTE/5G), WiFi, Bluetooth & BLE, Sub 1-GHz, Thread, WiSun, Zigbee LoRa, SigFox and Iridium Satellite. The Mini-PCIe slots provide customers with a flexible plug-n-play platform that can be tailored to meet their exact specifications. Different wireless standards support important communication modes, such as mesh and point to point networking.
A variety of sensor interfaces are available for monitoring data include camera, thermal, infrared, GPIO, USB, CAN, RS232, RS485, SPI and I2C (more discussed below).
One special feature of the Gateworks Single Board Computers is that they have a sleep functionality to conserve power. The Gateworks System Controller, or GSC, can be powered off of a coin cell battery. This allows vehicles that are parked for extended periods of time to conserve power. A timer can be used to occasionally wake the SBC up to read GPS coordinates and update the location to the cloud. Additionally, a vibration sensor can be connected to the SBC to wake the SBC up when the engine is started.
Single, dual and 5-port Gigabit Ethernet models are available including SFP Fiber support. These high speed wired connections are typically used to communicate with other onboard devices.
The Gateworks AI solution is capable of computing at the edge, whether that is storage, analysis or alerts. eMMC, MicroSD and mSATA storage expansion is available.
GPS is an optional soldered on option for Gateworks SBCs. A u.FL or MMCX antenna connector allows for a variety of antenna options. The typical GPS used on Gateworks SBCs supports NMEA, GLONASS, BeiDou, Galileo, QZSS, SBAS, WAAS/EGNOS/MSAS, Differential GPS (DGPS).
Artificial Intelligence Sensors & Data
Sensors play a crucial role in artificial intelligence. These may include camera data, GPS location, temperature, humidity, motion, vibration and infrared. Gateworks SBCs include GPIO, USB, CAN, RS232, RS485, MIPI, HDMI, SPI and I2C interfaces for connecting local sensors, including cameras.
Bluetooth Low Energy has become extremely popular for interfacing to vehicle sensors wirelessly. This sensor data can then be stored and analyzed locally on the Gateworks SBC or pushed up to a cloud server for more long term storage and heavier analysis.
A majority of vehicles utilize the CAN Bus for monitoring critical vehicle data, such as engine temperature, speed, and oil levels. SAE J1939 and SAE J2284 are common standards used in the automotive industry. Gateworks SBCs have CAN Bus capabilities.
AI Learning & Recognition
Gateworks SBCs can act as as platform for certain machine learning applications utilizing things such as a tensor processing unit (TPU) to detect machine abnormalities and patterns. Something such as an abnormal vibration may be a sign of a failed wheel bearing. Camera recognition can also be implemented through the TPU. The TPU device would interface through one of the high-speed Mini-PCIe slots. The detected information through the TPU can then be used to detect objects, phrases or abnormalities in data. Alerts can be dispatched based on detecting something specific.
A camera can detect and further learn different objects, whether it is counting people in crowds or detecting species of animals walking through the camera’s view.
AI Object Detection Demonstration
A demonstration has been created showcasing these two pieces working together in an image recognition scenario. The Google Coral Mini-PCIe card is inserted into a Gateworks SBC which also has an USB webcam connected to it. The Coral unit analyzes the live video from the webcam to detect and identify different objects. Each inference is extremely fast and accurate. A video of the demo is available.
Flexible Software & Clouds
All Gateworks Single Board Computerse support an Open Source Linux OS (Ubuntu, OpenWrt) which provides the user with complete access and control of the data and allows the use of any of the large number of cloud providers such as Microsoft Azure and Amazon Web Service (AWS). AI data can be sent to the preferred cloud of choice over a variety of connections, including cellular or satellite, for storage, analysis and alerts..
Gateworks can also provide customers with a “Special” configuration of our standard single board computers. This “Special” program allows customers to create cost and feature optimized versions of our standard products. The special is assigned a unique, revision locked, part number so customers can be insured they receive the same exact configuration every time they order. Note there is a 100-piece minimum order quantity for special configuration models. Some of the items that can be configured under the “Specials” program are:
- Changing components such as DRAM size, Flash size, or CPU type/speed
- Removing components not used to lower the cost of the product
- Adding optional components designed into the product but not loaded on the standard version (ie. locking headers)
- Programming the customer firmware
For high volume applications, Gateworks can also design a full custom SBC to meet the customer exact needs. Contact firstname.lastname@example.org for more information and to discuss your exact needs.
Recommended products for Artificial Intelligence
Recommended Single Board Computers:
- Newport GW6903 – Small 35mm x 100mm, 2x Half Mini-PCIe sites, 1x GbE, OcteonTx Dual Core CPU
- Newport GW6100 – Small 35mm x 100mm, 1x Mini-PCIe sites, 1x GbE, OcteonTx Dual Core CPU
- Newport GW6200 – Small 70mm x 100mm, 2x Mini-PCIe sites, 2x GbE, OcteonTx Dual Core CPU
- Newport GW6400 – 140mm x 100mm, 4x Mini-PCIe sites, 5x GbE, OcteonTx Dual Core CPU
- Venice GW7300 – 105mm x 100mm, 3x Mini-PCIe sites, 2x GbE, i.MX8M Mini Quad Core CPU
- Additional Venice and Newport SBCs available in a variety of form factors and with up to 4x Mini-PCIe sites. See here for more information.
Recommended TPUs for Artificial Intelligence:
- Google Coral TPU – Half Size Mini-PCIe Tensor Processing Unit
Recommended Mini-PCIe Cards for Wireless Support:
- GW16122 Sub-1GHz Radio – Based on Ti CC1352P Multiprotocol MCU
- GW16132 LTE CATM1 Cellular & BLE 5.0 Radio – Based on ublox SARA-R410 CATM1 Cellular Module and NINA-B301 BLE module (Nordic nRF52840 chipset)
- GW16130 Iridium SBD Satellite Radio – Based on Iridium 9603N SBD Module
- GW17032 802.11ac/abgn WiFi Radio – Based on Qualcomm QCA9880 Chipset, 3×3 MIMO
- GW17039 802.11bgn+BLE 4.0 Radio – Based on Qualcomm AR9565 Chipset
- GW17046 LTE CAT4 Cellular Radio – Based on ublox Toby-L2 Module
- Various 5G Cellular Modems
- Additional 3rd Party Radios are Available for Support of other Wireless Standards such as LoRa and SigFox.
- Newport Family – Indoor Aluminum Enclosure
- Additional 3rd Party Enclosures are available for Hazardous Area and Outdoor Installations.