Overview
On October 23, during WTCC 2026 in Shanghai, Arm China’s Gitee (Jishu) community hosted the Jishu Maker Workshop on small embodied-AI solutions, working with open-source AIOS alliance partners Rockchip, Radxa and RT-Thread to showcase on-device AI and formally launch the first Jishu-series deliverables—JishuFriends and JishuBuddy. The first JishuFriends prototype, Puppy, is a robot dog integrating environmental perception, voice-vision interaction and autonomous movement, closing the perceive-to-act loop. JishuBuddy is a next-generation AI hardware development tool for Agentic AI, using natural language as the entry point and targeting Arm-based embedded devices.
Background & Interpretation
1. Technology Background and Evolution
Arm China launched the open-source AIOS alliance in July, focusing on persistent perception, autonomous execution and long-term memory, and rallying domestic AI academia-industry partners to build next-generation AI infrastructure. The industry is pushing perceive–decide–act capabilities down to power- and cost-constrained embedded devices, moving Physical AI and Agentic AI from demos into real scenarios.
2. Core Drivers and Underlying Mechanisms
LLM inference is shifting from compute-intensive to memory-intensive, making bandwidth the on-device bottleneck. Rockchip proposes decoupling compute from control via a main SoC plus AI co-processor to lay the compute foundation for small embodied devices. Radxa’s MicroDuck with the ZERO 3W helps third parties prototype quickly on open hardware; RT-Thread demonstrates three Agent schemes—robot dog, HyperCar, Codex orchestration—pushing agents from conversation to execution.
3. Market Structure and Industry Chain Effects
On-device embodied AI is competing less on model parameters and more on software-hardware co-design and developer ecosystems. Arm China’s booth—displaying wafers and bare dies to trace a chip’s birth—highlights its full-stack positioning from IP to dev tools. The open-source AIOS alliance pulls chipmakers, hardware firms and OS vendors into one collaboration network, lowering the barrier for smaller developers and likely spawning more small but sharp on-device applications.
Implications & Outlook
Near term, the Jishu series gives developers a toolchain template from prototype to production, accelerating consumer products like on-device robot dogs and smart hardware. Medium term, as on-device compute and memory bandwidth improve, Agentic AI will extend from phones and cars to robots and IoT, and natural-language programming will further lower hardware-development thresholds.
Takeaways for Industry Participants
- Chip and module vendors can optimize storage and interconnect around the SoC plus AI co-processor architecture to attack the on-device bandwidth bottleneck.
- Hardware startups should leverage open hardware and alliance ecosystems to shorten the sample-to-volume cycle.
- Developers can use JishuBuddy to build AI apps via natural language, turning ideas into prototypes faster.
This column compiles industry information and shares technical perspectives; it does not constitute investment advice.