At the DDC 2025 conference, D-Robotics announced a slate of upgrades to its development infrastructure and introduced the RDK S600, a compute platform built for embodied intelligence. The S600 is expected to launch in the first quarter of 2026.
The company also unveiled a development platform that integrates three core capabilities:
- First, it includes a data closed-loop system that links models, simulation, and hardware to support autonomous system iteration.
- Second, it provides an embodied intelligence training ground that offers full-chain task support across varied scenarios.
- Third, it introduces agentic development tools that enable developers to build and deploy robotic applications using natural language.
As robotics move into more complex environments, neither hardware nor software alone can meet performance demands. Increasing task complexity and more advanced models have intensified the gap between on-device compute limitations and real-time requirements. Cloud resources can offset those limits, which makes edge-cloud collaboration a natural progression. Rising competition is also putting pressure on development speed. Hardware and software co-design and integrated architectures improve performance while shortening the cycle from R&D to deployment.
To address both compute constraints and development efficiency, D-Robotics has built a full-chain system spanning device and cloud. On the device side, the company continues to refine its BPU (Bernoulli processing unit) architecture and has developed two product lines for different compute tiers. On the cloud side, its platform provides hundreds of ready-to-deploy robotics algorithms to help teams build competitive solutions more efficiently.
D-Robotics also showcased the S600, which delivers 560 TOPS (tera operations per second) of compute (at INT8 precision) and supports edge deployment of large models. When running models such as Pi-Zero and Qwen2.5-VL-7B, its performance reportedly exceeds mainstream platforms by more than 2.2 times, meeting the decision-making demands of real-world environments.
In interviews with 36Kr and other outlets, CEO Wang Cong said the company does not intend to deliver turnkey end products. Instead, it focuses on identifying common technical needs across scenarios and refining them into standardized modules. Stereo vision, for example, is used in categories spanning robot vacuums, lawn mowers, drones, and quadruped robots. By targeting such bottlenecks, D-Robotics aims to support a wide range of product requirements. In embodied intelligence, its motion control algorithms are optimized for adaptation and generalization.
“For R&D needs, we want to turn common processes such as data collection, labeling, generation, simulation, and testing into standardized toolboxes,” Wang said. “Like PowerPoint, we don’t interfere with creativity. We provide robust foundational tools that support highly abstract, multi-scenario R&D workflows.”
D-Robotics is prioritizing three deployment areas: mass produced robotic products, emerging robotics applications, and future general-purpose embodied robots.
According to 36Kr, the company has partnered with more than 60 players across the robotics supply chain. These include supporting Narwal on the Xiaoyao 002 robot vacuum and bringing AI-powered stereo perception to the category; enabling Insta360’s launch of what is billed as the world’s first panoramic drone, the Antigravity A1; and working with Vbot on a quadruped companion robot.
Over the past year, D-Robotics said product shipments rose roughly 180% year-on-year, while its customer base grew about 200%.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This article was written by Huang Nan for 36Kr.