← Back to Products

TSING-ONE

Model TSING-COM-B · Embodied Intelligence Edge Compute Unit

Qingfang — embodied intelligence hardware that thinks and acts

TSING-ONE runs the CogniBrain OS, and can be equipped with a 14-DOF humanoid dexterous hand. It supports natural-language task understanding and long-horizon autonomous planning, adapts to different robot bodies, and serves manufacturing, warehousing, logistics and service-collaboration scenarios.

SYS / PROD 04
MODEL / TSING-ONE
REV 2026.08

Built for real industrial sites

One Controller, Any Brand

Natively adapts to FANUC, KUKA, ABB, UR and other mainstream industrial robots, with full industrial protocols — Modbus, Profinet, EtherNet/IP — for unified cross-brand scheduling with zero redevelopment.

On-device LLM, Smart Offline

≥ 100-150 TOPS INT8 of total AI compute runs 7B-70B parameter models on-device with <200ms end-to-end response — full intelligent decision-making even with no network.

Dual-path, Secure & Compliant

Offers both international and domestic trusted solutions — Phytium / Hygon CPUs with Cambricon NPU — meeting industrial security compliance and domestic-substitution (Xinchuang) requirements.

Plug & Play, Live in 1 Day

Standardized hardware interfaces + open APIs, compatible with existing production lines, security systems and ERP/OA — from unboxing to operation in one day.

The full stack, from perception to action

Cognitive Brain Driven

Powered by CogniBrain OS — natural-language task understanding, intent inference and long-horizon multi-task planning. It understands instructions, not fixed actions.

Dexterous Operation

14-DOF humanoid dexterous hand with fingertip tactile sensing array; precision grasping repeatability of ±0.5mm for fine assembly and flexible sorting.

Multi-modal Perception

6 RGB cameras + LiDAR + full-body IMU array for millisecond scene understanding and real-time 3D semantic map reconstruction — perception is decision.

On-device Real-time Inference

≥ 100-150 TOPS INT8 edge compute completes the perception-reasoning-control loop without the cloud, for low-latency response in network-free industrial environments.

Force-controlled Safety

Full-body compliant force control with <5ms collision detection and <2cm safe-stop distance; ISO 10218 certified for zero-risk human-robot collaboration.

Rapid Hardware Adaptation

Standardized HAL hardware abstraction layer decouples software from the body — humanoid, wheeled and cobot platforms adapt quickly without retraining.

Products already adapted on CogniBrain OS

Wheel-legged Robot photo

Wheel-legged Robot

ADAPTED / OK

Dexterous Operation Robot photo

Dexterous Operation Robot

ADAPTED / OK

Quadruped Robot photo

Quadruped Robot

ADAPTED / OK

Humanoid Torso photo

Humanoid Torso

ADAPTED / OK

PHOTO CREDIT / CSIRO (CC BY 3.0) · P. MASSEY / NOIRLAB (CC BY 4.0) · USAF · US NAVY (PD) · WIKIMEDIA COMMONS

Multi-industry deployment · full-scenario coverage

Automotive scene

Automotive

Stamping handling, bolted assembly and weld-point QC scanning — built for ultra-wide, three-side continuous production lines.
Warehouse & Logistics scene

Warehouse & Logistics

Multi-SKU precise sorting, cross-level transport and shelf tidying — 3-5x efficiency over conventional operations.
3C Electronics scene

3C Electronics

Precision component inspection, flexible PCB sorting and AOI assisted re-check — higher yield, fully traceable.
Service & Collaboration scene

Service & Collaboration

Medical delivery, smart-building wayfinding and front-desk reception — open-scenario tasks via natural voice interaction.

PHOTO CREDIT / S. JURVETSON (CC BY 2.0) · OXFAM EAST AFRICA (CC BY 2.0) · WIKIMEDIA COMMONS

Specifications

ROUTE A

International Technical Route · AIPC Hardware Specifications

Compute SOC & Unified Memory
CPU16 cores / 32 threads, base clock ≥ 3.5GHz, boost ≥ 5.0GHz, L3 cache ≥ 32MB; AVX2 and matrix / vector extension support; hardware virtualization and IOMMU.
GPUIntegrated general-purpose compute GPU with FP16/BF16/INT8 matrix operations; peak FP16 compute ≥ 20 TFLOPS; VRAM bandwidth matched to unified memory bandwidth.
NPUDedicated AI accelerator, INT8 compute ≥ 40-80 TOPS; combined CPU + GPU + NPU AI compute ≥ 100-150 TOPS INT8.
MemoryOnboard LPDDR5X, 8 channels, theoretical bandwidth ≥ 256GB/s.
Wired LAN1× 2.5GbE + 1× 10GbE (RJ45 or SFP+), VLAN / QOS support.
WirelessWiFi 6E or WiFi 7, Bluetooth 5.3 and above.
Industrial & Robotics / Sensor Support
Serial / IO dockUSB3 / USB4 · 4× RS-485/RS-232, 8-16 DI/DO, 4-8 analog channels.
Sensor gatewayGigabit Ethernet · RS-485 / 4-20mA / IO-Link → Modbus TCP / OPC UA.
Robot controller linkHost Gigabit Ethernet · FANUC FOCAS / EtherNet-IP, KUKA EKI, ABB OPC UA.
Industrial / 3D camerasHost NIC + USB3 · multi-channel GigE Vision and USB3 Vision; camera PoE from switch.

Get robots truly working on your line

Apply to test TSING-ONE and explore cross-brand scheduling and on-device intelligence.

Apply for a Test
SYS / ACTIONREV 2026.08