Qualcomm introduced a high-end hardware development kit called the Robotics RB6 Platform that can act as the brains of autonomous robots, industrial drones, and delivery robots.
As artificial intelligence becomes a driving force for robots, so has the hardware under the hood powering it. Qualcomm said it upgraded the RB6 system with its Cloud AI 100, which is designed for AI inferencing in data centers, giving it more than 10X the performance of its predecessor released in 2020, the RB5. It adds state-of-the-art 5G connectivity as well, with support for ultra-fast millimeter-wave 5G networks.
Dev Singh, head of autonomous robotics, drones, and intelligent machines, said the unique combination of AI and 5G abilities in Qualcomm’s RB6 would let its customers build smarter, safer, and more advanced robots.
Qualcomm CEO Cristiano Amon, who ran its mobile-chip business before getting promoted last year, has pushed the company to diversify its offerings as 5G advances into areas besides smartphones. It is using its expertise in engineering chips for smartphones—where battery life is a key priority—to win over more customers in the world of autonomous robots and cars.
At the start of the year, GM said that it would use a cluster of Snapdragon Ride chips designed by Qualcomm to serve as the brains of its Ultra Cruise automated driving system that is due to arrive in vehicles by 2023.
Server-Class AI
Inside the RB6 system is Qualcomm’s QRB5165, a robotics-grade system-on-a-chip (SoC) manufactured on the 7-nm process. At the heart of the processor is the Kryo 585 CPU, the same that sits inside its Snapdragon 865 for smartphones.
Clocked at frequencies of up to 2.84 GHz, the CPU uses a combination of four 64-bit high-performance cores that run hefty computational workloads and four high-efficiency cores that help prolong a robot’s battery life.
The processor also contains Qualcomm’s Hexagon DSP 698 with Hexagon Vector Extensions (HVX) and Tensor and Scalar acceleration engines to pump out up to 15 trillion operations per second (TOPS) for on-device AI and deep-learning workloads without burning through excess battery life. The chip also has hardware acceleration for advanced computer vision (CV) with its Engine for Video Analytics (EVA) block.
Qualcomm said the AI engine inside the SoC is supplemented by the Cloud AI 100 accelerator. This gives the RB6 many times more processing power for AI workloads than the RB5, which is also based on the QRB5165.
Leveraging the Cloud AI 100, the RB6 system can supply from 70 up to 200 TOPS of performance.
The RB6 platform also packs in more vision-processing prowess than the RB5 to help robots sense and react to obstacles in the surrounding world. The system connects to 12 cameras via D-PHY and 18 cameras via C-PHY (up to seven concurrently). In addition, it can process up to 24 1080p video feeds at the same time. Also under the hood is its Adreno 650 Visual Processing Subsystem, which unites an Adreno GPU and a video processing unit (VPU).
Secure Networking
Backing everything up is a secure processing unit (SPU) that acts as a secure vault in hardware to safeguard data such as faces, iris scans, and other biometric information that must remain secret. The SPU is designed to deliver features including hardware root of trust, a trusted execution environment, as well as secure boot.
Designed with a dual M.2 card form factor, the RB6 platform is based on a flexible architecture that accepts expansion cards. These cards can be used to add support for Release 15, and Releases 16, 17, and 18 features to 5G networks when they are available.
Designed to handle harsh industrial environments and operate in temperature ranges of −30 to 105°C, the QRB5165 can communicate over industrial protocols such as EtherCAT and time-sensitive networking (TSN).
The dual 14-bit Spectra 480 image signal processor (ISP) handles up to two trillion pixels a second with high-performance capture of 200-Mpixel images, 8K video recording, and 4K video capture.
Qualcomm said the QRB5165 can be supplemented with 16 GB of LPDDR5 DRAM running at up to 2750 MHz.
The RB6 development kit adds advanced connectivity using Qualcomm’s FastConnect 6800 System with Wi-Fi 6.
Fierce Competition
Building chips for autonomous robots and embedded devices has become big business for many vendors.
Qualcomm is also supplying several of the chips that serve as the brains of Amazon’s Astro, an autonomous home robot powered by Alexa that uses a suite of cameras and other sensors to model and navigate a room. The company has said that its RB3 platform is in use by hundreds of customers, including iRobot.
NVIDIA is several generations into its “Jetson” family of energy-efficient computer modules for autonomous vehicles and robotics. At the heart of its latest Jetson is the “Orin” SoC, which is based on its Ampere GPU architecture. Orin pumps out 6X more performance than its previous generation for a total of 200 TOPS, offering what CEO Jensen Huang calls a server’s worth of performance that can fit in the palm of your hand.
Agricultural equipment giant John Deere is in the final phases of building its first fully autonomous tractor, which sports six pairs of stereo cameras and a suite of other sensors powered by NVIDIA’s Jetson to drive itself.
NVIDIA has also started offering its Isaac software development kit (SDK), a suite of tools that helps customers program robots based on its hardware. On top of that, it rolled out Isaac Sim, a robot simulation environment powered by Omniverse.
Robotic Blueprints
Qualcomm said the RB6 is backed by its Intelligent Multimedia SDK, a set of software tools comprised of AI, computer vision, as well as networking and multimedia building blocks to support end-to-end deployment of robotics. RB6 supports a range of industrial middleware, including the Robot Operating System (ROS).
It also announced a new RB5 reference design that comes with all of the hardware and sensors for building autonomous mobile robots (AMRs) and is based on the RB5 platform deployed in millions of systems.