As the current AI wave evolves from generative AI to physical AI capable of interacting with the real world, Qualcomm unveiled its new high-end processor, the Dragonwing IQ10, specifically designed for the robotics field. This chip not only inherits the computing advantages of Snapdragon but also incorporates the PC-grade Oryon CPU architecture, aiming to provide a powerful "brain" for humanoids, autonomous mobile robots (AMRs), and industrial robots.
Dragonwing IQ10: Balancing Performance and Security
According to data released by Qualcomm, the Dragonwing IQ10 uses a Qualcomm Oryon CPU design with up to 18 cores, showing that Qualcomm is bringing its proven architectural advantages in the PC field (Snapdragon X Elite) to the robotics field to provide optimized performance for both single-threaded and multi-threaded applications.
In terms of AI computing, the Dragonwing IQ10 is equipped with a dedicated Hexagon NPU designed for Visual Language Action Models (VLA), with a maximum AI computing power of 700 TOPS (in sparse computing mode). This computing power configuration is mainly to meet the robot's needs for sensor pre-processing, post-processing, and complex computer vision (CV) computations.
Addressing the critical needs of "vision capabilities" and "safety" in industrial settings, the Dragonwing IQ10 supports the simultaneous operation of up to 20 cameras, which is essential for industrial robots requiring ambient awareness and precision operation. Furthermore, the chip incorporates a Real-time Safety Subsystem that meets SIL3 (Safety Integrity Level 3) standards, ensuring the safety of robots when collaborating with humans.
Unified Architecture: From Chips to Skills
Beyond hardware specifications, Qualcomm places greater emphasis on integrated hardware and software solutions. Qualcomm has proposed a unified "Silicon-to-Skills" architecture to address the current fragmentation in robotics development.
This architecture encompasses everything from the underlying SoC and Qualcomm Linux/Ubuntu operating system to the upper-level "Compound AI Robotics SDK." Qualcomm hopes that this platform will enable developers to more easily deploy various robotic "skills," such as pick and place, tool manipulation, object detection and tracking, and path planning.
In addition, Qualcomm has also proposed the concept of Physical AI MLOps (Physical AI Machine Learning Processes) for backend infrastructure. Through virtual platforms and digital twin technology, data augmentation and benchmarking are performed to accelerate the learning curve of robots.
The vision for the application of AI in the real world: to solve industry pain points
Qualcomm believes that future robots will no longer be merely automated devices executing single commands, but rather "physical AI" carriers capable of continuous learning. Through heterogeneous edge computing and hardware-software co-design, robots will possess the capability of mixed criticality AI.
In practical applications, Qualcomm focuses on three major vertical sectors: logistics, manufacturing, and retail. For example, in the retail industry, robots can perform tasks such as shelf restocking and individual item identification; in the manufacturing industry, they can assist with circuit assembly and parts distribution.
Qualcomm has already planned a general-purpose robot (GPR) reference design based on the Dragonwing IQ9, and also provides reference designs for humanoid robots, thereby accelerating the product launch of its ecosystem partners.
Qualcomm's Robotics Market Strategy Upgrade
Qualcomm's launch of the Dragonwing IQ10 series processors, distinct from its previous Robotics RB series, symbolizes a strategic upgrade in the robotics field.
As discussions about humanoid robots like the Tesla Optimus continue to gain momentum, the market demand for computing power in "embedded AI" has exploded. Qualcomm's latest offering of an 18-core Orion CPU and a 700 TOPS NPU computing power is clearly aimed at challenging NVIDIA's dominant position in the robotics chip field (such as Jetson Thor).
The advantage of the Dragonwing IQ10 lies in Qualcomm's long-term technological accumulation in low power consumption, communication (5G/Wi-Fi), and edge computing. If Qualcomm can successfully establish a developer ecosystem similar to "Silicon-to-Skills," it will have the opportunity to accelerate the entry of robots from factories into homes and commercial settings, realizing the vision of "any entity can become a continuously learning robot."
