Qualcomm today launched the Robotics 5G Platform, which it claims is the world’s first to support AI acceleration and 5G. It occupies the high end of the chipmaker’s prefab robotics solutions and is designed to enable things like machine learning, heterogeneous computing, and computer vision while withstanding “industrial-grade” temperatures and incorporating security “at every layer.”
Stakeholders like Qualcomm believe AI is moving to the edge. TrendForce predicts the edge computing segment of products will grow by a compound annual growth rate of more than 30% from 2018 to 2022. According to Forrester, the edge cloud service market will expand by at least 50% in 2020 as providers deliver new infrastructure-as-a-service and cloud-native programming products on distributed devices.
“[W]ith billions of connected [devices communicating] and the massive amount of data that’s being enabled, robotics is one of the beneficiaries [of 5G and AI] — it’s going to benefit from these technologies,” Qualcomm senior director Dev Singh told VentureBeat in a phone interview. “We think that [5G and AI] are going to be really, really elementary for future robotics.”
Qualcomm intends to tap into the trend with the RB5 Platform, which builds on the RB3 Platform it launched early last year. Like the RB3, the RB5 comprises a set of hardware, software, and development tools tailored for consumer, enterprise, defense, industrial, and professional use cases. It packs a Qualcomm QRB5165 processor tailored for robotics applications, with a heterogeneous computing architecture that’s paired with Qualcomm’s fifth-generation AI Engine. An image signal processor (ISP) supports up to seven concurrent cameras to facilitate simultaneous localization and mapping (SLAM), object detection and classification, autonomous navigation, and path planning. A dedicated engine for video analytics plus Qualcomm’s Hexagon Tensor Accelerator and a companion modem that works with both 4G and 5G networks, and native support for the EtherCAT standard, an ethernet-based industrial protocol with integrated safety mechanisms.
“[5G] is going to drive the [next] industrial revolution — it’s going to be the connectivity fabric for the factory of the future,” Singh said. “5G is the first wireless technology that can handle the latency of real time … and it brings in things that replace a lot of the traditional communication protocols. 4G can do a lot, but it’s not as reliable — you can have failures, for example like link failures. 5G delivers more bang for your buck, especially for mission-critical services.”
A bit more on the RB5’s QRB5165: It contains an octa-core Qualcomm Kryo 858 paired with an Adreno 650 GPU and discrete digital signal processors for compute, audio, and sensors. The ISP — the Spectra 380 — can process 2 gigapixels per second and features Dolby Vision video capture, 8K video recording at 30 frames per second, 200-megapixel photos, and simultaneous capture of 4K HDR video at 120 frames per second and 64MP photos with zero shutter
lag. Another chip — Qualcomm’s Secure Processing Unit (SPU) — includes secure boot, cryptographic accelerators, malware protection, secure token for remote attestation and secure device provisioning, and biometric authentication (including fingerprint, iris, voice, and face).
According to Qualcomm, the QRB5165 — along with the fifth-gen AI Engine — delivers up to 15 TOPS of AI performance (TOPs refers to tera operations per second). That’s compared with the RB3’s 3 TOPS.
In developing RB5, Qualcomm entered into a strategic collaboration with TDK, the Japanese multinational electronics company that manufactures electronic materials, components, and recording and data storage media. TDK supplied the six-axis inertial measurement unit, bottom-port microphone, and robotic application algorithms within the RB5, as well as the Hall effect, temperature, and barometric pressure sensors.
RB5 is part of the RB5 Development Kit, a customizable platform offering support for Linux, Ubuntu, and robotics middleware like Robot Operating System (ROS) 2.0 and pre-integrated drivers for cameras, sensors, and 5G connectivity. It’s compatible with OpenCL, OpenGL, and OpenCV, in addition to depth-sensing cameras like Intel’s RealSense Depth Camera D435i and Panasonic’s TOF Camera. And it comes in flavors packing TDK’s ultrasonic time-of-flight and embedded motor control solutions built to withstand commercial and industrial-grade temperature ranges.
Commercial products based on the RB5 Platform are expected to arrive in 2020, but a development kit is available for purchase now from Thundercomm. (Customers will be able to opt for extended lifecycle support until 2029.) Qualcomm says over 20 early adopters are in the process of evaluating the platform and more than 30 of the company’s ecosystem partners are developing support hardware and software to enable robotics applications, including AirMap, Fractal.ai, Canonical, and ModalAI.