Contact Us
HARDWARE · ROBOT COMPUTING PLATFORM

Robot Core Controller — Brain and Cerebellum Across Every Compute Tier

A robot computing platform family covering brain and cerebellum: from production-ready RQX-59G (AGX Orin) and Robo3588 (RK3588 + RK1828 + TC397) to the embodied-AI flagship RQX-71G (Jetson Thor). GMSL2 automotive video links across the family, customizable per project.

6 → 2070 TFLOPS
Every compute tier in one family
12 ch
GMSL2 automotive camera inputs (max)
5 models
Production-ready and in-development flagships · Transparent roadmap

Core Capabilities

Brain + cerebellum · Integrated hardware-software · Automotive video links

01

Every Compute Tier

6 TOPS (RK3588) → 26 TOPS (+RK1828) → 100 TOPS (Orin NX / IQ9) → 200 TOPS (AGX Orin) → 2070 FP4 TFLOPS (Jetson Thor), selected by scenario and cost tier.

02

GMSL2 Automotive Cameras

Up to 12 channels on the in-house IQ9, 8 on RQX-71G and up to 8 on RQX-31G; coaxial links carry video, control and power (PoC) with multi-camera frame sync and hardware timestamps.

03

Brain + Cerebellum

RQX-71G / 59G act as the perception and decision brain (VLA / VLM models), while Robo3588 serves as the real-time control cerebellum — one combination covering the full embodied-AI chain.

04

Real-Time Control & Motion

EtherCAT real-time Ethernet (2× 100M dedicated ports / gigabit support), up to 6 CAN FD channels, DI / DO and PWM for multi-joint synchronization and highly dynamic motion.

05

Safety Islands & Reliability

Two safety-island routes: IQ9's built-in 4×Cortex-R52 real-time cores, or the TC397 (with HSM) on RK platforms; -40~85°C wide temperature with CE / FCC certifications.

06

Integrated Hardware-Software

Pre-installed or optional ACOS middleware, DDS-SOA communication, TSN solutions and ADrive algorithms; native ROS 2 support for both RoboVan and AI Box scenarios.

Model Specifications

RQX-71G
NVIDIA Jetson Thor T5000 / T4000 · 2070 / 1200 FP4 TFLOPS · humanoid robotics and local VLA / VLM inference · 8× GMSL2 + QSFP28
RQX-59G Series
NVIDIA Jetson AGX Orin 32GB / 64GB · 200 / 275 TOPS (INT8) · in production · multi-sensor fusion and low-speed autonomy · -E model adds a PCIe expansion box
RQX-31G
NVIDIA Jetson Orin NX 16GB · 100 TOPS (70 / 40 optional) · compact ROS 2 controller · 9-36V wide input with ignition control · VESA mounting
Robo3588
RK3588 + RK1828 + AURIX TC397 · approx. 26 TOPS INT8 · in production · runs 7B-class LLMs locally · dual gigabit with EtherCAT / TSN
In-House IQ9 Controller
Qualcomm QCS9075 with 4×Cortex-R52 safety island · 100 TOPS dense / 200 TOPS sparse · 12× GMSL2 + automotive Ethernet · in development (automotive target -40~85°C)
Storage & Expansion
DDR5 / LPDDR5 8–128GB; NVMe / eMMC / M.2 expansion; optional RK1820 compute card or SSD
Network & Positioning
5G / 4G, Wi-Fi 6 / BT 5.0 (per model); external GNSS / IMU modules recommended for positioning
Software Ecosystem
Ubuntu / JetPack / Android; native ROS 2 support; stacks with ACOS, DDS-SOA, TSN and ADrive for integrated solutions

Use Cases

Humanoid & Embodied AI

RQX-71G runs VLA / VLM models locally as the embodied-AI compute foundation.

AMR / AGV & Mobile Robots

RQX-59G / 31G with multi-sensor fusion and the ROS 2 ecosystem, from logistics to research and education.

RoboVan & AI Box

Low-speed functional vehicles, unmanned vehicles and edge AI boxes with Robo3588 / RQX-59G selected on demand.

Related Products & Solutions