The surging rise of robotics is driving a full-chain technological revolution from sensors to AI (Artificial Intelligence).
Wayne Dai, Chief Strategy Officer and General Manager of the IP Business Unit at VeriSilicon, pointed out at the "VeriSilicon Embodied Robot Special Technology Seminar" that a robot is essentially a mobile AI machine. Its realization relies not only on computing power but also requires the deep integration of CPUs, GPUs, and AI computing, along with efficient task collaboration between the cloud and the edge tailored to specific scenarios.
Figure | Wayne Dai, Chief Strategy Officer and General Manager of the IP Business Unit at VeriSilicon
Throughout this technological transformation, low latency and low power consumption remain core challenges. How can request interactions with the cloud be completed within an extremely short response time? Wayne Dai suggests that a viable direction is to develop low-latency technology with an ultra-high compression ratio to reduce reliance on high-bandwidth memory, thereby ensuring real-time performance while lowering power consumption. This approach has been validated in fields such as wearable devices and is equally applicable to robotic systems with stringent sensor data processing requirements.
As the key to realizing the "eyes of robots," visual processing technology is a field VeriSilicon has deeply cultivated for a long time. From automobiles to robots, the difficulty of environmental perception has significantly increased; it requires not only clear vision but also the rapid processing of complex and dynamic lighting conditions. VeriSilicon's Image Signal Processor (ISP) and Display Processor provide targeted solutions to these challenges: in extreme environments with strong light and glare, algorithms such as sunlight readability enhancement and local tone mapping ensure images remain clear and distinguishable. Meanwhile, for service robots emphasizing interactive expressiveness, high-refresh-rate and low-latency GPU interconnection technologies guarantee smooth and delicate animations.
VeriSilicon has distilled these underlying technologies into six major categories of self-developed and controllable processor IPs, covering the full chain from graphics and neural networks to video and digital signal processing. These IP modules can be flexibly scaled and combined according to performance and functional requirements. Furthermore, they have been deployed in products from leading customers in fields such as cleaning robots, home companion robots, and drones, forming a market-validated technology foundation.
Moreover, the engineering implementation of these technologies is equally critical. VeriSilicon provides not only IP but also transforms its extensive accumulated experience in the automotive and robotics fields into reusable solutions through a chip customization platform that integrates software and hardware. Coupled with AI-oriented API framework standards, this approach helps customers develop edge AI applications more efficiently.
From the precise perception of sensors to the efficient computing of AI, VeriSilicon's technology layout spans the entire process of robots evolving from "seeing" to "thinking." Wayne Dai emphasized that this system-level capability, spanning from sensors to AI, is the core cornerstone supporting the maturation of future diverse robotic devices.