D-Robotics Debuts Full-Stack Industrial AI for Vision, Data, and Embodied Robotics at CIIF
The 26th China International Industry Fair opened on Oct. 12 at the National Exhibition and Convention Center (Shanghai). D-Robotics made its first appearance at the event, bringing its Sunrise intelligent computing chips, RDK robotics developer kits, and RoboGo software platform to showcase full-stack AI capabilities for visual inspection, industrial data collection, and industrial embodied AI. The fair is a major showcase for industrial automation and robotics.
For industrial customers, D-Robotics integrates edge computing, supporting algorithms, and cloud development tools. Its self-developed BPU AI computing architecture and product lineup from 5 to 560 TOPS address varied compute demands, while more than 30 open-source algorithms and 200-plus application examples reduce duplicated development in hardware selection, model adaptation, and tool integration.
The company says cumulative Sunrise chip shipments have exceeded 8 million units, with customer coverage in embodied AI surpassing 50% and mass-production validation from more than 400 robotics customers worldwide. D-Robotics is extending its volume-production experience from consumer segments such as lawn mowing, cleaning, and companion robots into industrial applications including visual inspection, industrial domain control, and embodied robots.
In industrial vision, D-Robotics offers computing solutions based on the Sunrise series and works with ecosystem partners including Quectel, Tuwu Technology, and Runxin Micro to launch standardized AI vision boxes. These products have been deployed at scale in color sorting, quality inspection, and sorting tasks. The standardized boxes aim to simplify deployment for manufacturers.
Industrial AI development requires continuous use of on-site data. How to process data, train and iterate models in a company-controlled environment, and then deploy results to equipment remains a practical challenge for data-collection firms and integrators.
RoboGo is D-Robotics' software platform for embodied intelligence and intelligent terminals, covering data acquisition, model training, simulation validation, and quantized deployment. It supports one-click deployment of cloud-trained results to edge devices, providing a unified development environment for industrial data services, production-line proof-of-concept work, and subsequent model iteration.
To meet industrial data management needs, RoboGo supports private deployment. D-Robotics says it has completed such deployments for data-collection companies including Fanyu Zhiyuan, allowing data processing and model training within corporate private environments and meeting the requirement that industrial data does not leave the factory. This approach is intended to address data-security and compliance concerns.
In concrete collaboration, Fanyu Zhiyuan handles data processing and human-in-the-loop work; Yingmou and Mouxianfei convert real-world scenarios into simulation environments for training; and Xingyi Intelligent connects data supply with model services. By linking data, simulation, and training, RoboGo helps customers use on-site data for algorithm iteration and deploy models to edge devices.
Industrial embodied robots must perceive their environment while completing positioning, movement, and manipulation. For tasks such as flexible wire-harness assembly, auto-parts sorting, and machine-tool loading and unloading, edge computing platforms need to run multiple model types and coordinate with robot body control.
At the fair, D-Robotics highlighted high-compute products such as the Sunrise S100 and S600 for these scenarios. The Sunrise S600 has served more than 20 embodied robotics companies, including Xiaoyu Zhizao, UBTECH, Tashi Zhihang, Beijing Humanoid Innovation Center, and Shanghai Humanoid, covering industrial welding, flexible wire-harness assembly, power inspection, auto-parts sorting, machine-tool loading and unloading, and logistics handling. Most projects have entered mass-production-level shipments.
Beyond compute, D-Robotics continues to develop proprietary algorithms including 3D spatial perception, VO-DP+ dexterous manipulation, HoloBrain general manipulation foundation model, MultiCAM SLAM omnidirectional multi-camera positioning, and WM-LOCO world-model-driven obstacle negotiation.
These algorithms cover key robot capabilities in perception, positioning, movement, and manipulation, providing algorithmic support for complex industrial task development. The company says these capabilities are intended to reduce development effort for complex industrial tasks.