Applications

The whole robot, from sensor to decision.

Embodied AI needs more than a brain. YUAN's full-stack robotics platform — Holoscan Sensor Bridge capture, Jetson-class compute, and the NexVDO SDK — turns raw sensor streams into behaviour.

2070: FP4 TFLOPS on ARC6N0 T5X, built on NVIDIA Jetson Thor
275: TOPS on the Arc AGX robotics platform
6: robotic forms served, from humanoids to AMRs

The brief

Six robotic forms, one integration problem.

Embodied AI is arriving in every shape at once: industrial arms holding tolerance on complex factory lines, humanoids coordinating perception with movement, quadrupeds adapting to terrain on patrol, drones flying high-resolution inspection routes, AMRs mapping warehouses, and service robots greeting people at the door.

Each form asks the same hard question — how do you bridge bottom-layer sensor integration with top-layer edge AI computing? A robot's cameras, depth sensors, LiDAR, radar, IMU, GNSS, and RTK all speak different languages, and its intelligence is only as good as the pipeline that carries them.

YUAN answers with a full-stack hardware platform and one SDK, so robotics builders integrate once and deploy across every form.

Six robotic forms, one integration problem.

The software

NexVDO SDK: five verbs for autonomy.

Capture moves data and video into the system with an RDMA pipeline and PTP-synchronised sensors, so every stream shares one clock. Record encodes in H.26x or AV1 with AES-secured storage and metadata packaging — fail-safe reliability for mission-critical inspection footage.

Stream carries video over ONVIF, RTSP, and WebRTC for low-latency teleoperation, keeping a remote operator's hands effectively on the machine. Analyze runs edge inference — vision models, ASR and TTS speech processing, and support for large language and vision-language models.

Automate closes the loop: navigation, mission planning, and localisation with native ROS2, MAVLink, Visual Localization, and SLAM and NAV2 integration. Five verbs, one toolkit, no glue code.

The hardware

From sensor bridge to safety island.

The platform natively integrates cameras, depth cameras, LiDAR, radar, IMU, GNSS, RTK, and industrial protocols like EtherCAT and motor control. Holoscan Sensor Bridge captures the high-bandwidth streams a robot's perception modules produce — 10G/4K, 10G/HD — and feeds them through a Functional Safety Island directly to the main computational node.

That node scales with the mission. The Arc AGX platform, powered by NVIDIA Jetson, delivers 275 TOPS in a highly integrated core; the ARC6N0 T5X, built on NVIDIA Jetson Thor, reaches 2070 FP4 TFLOPS for the heaviest multi-modal workloads.

The design goal is unglamorous and absolute: zero-latency streaming and high system reliability, because a robot that hesitates is a robot that fails.

In the field

DOGU's patrol robots run on YUAN.

DOGU Robotics integrates YUAN's VPP6N0-S AGX platform, built on NVIDIA Jetson AGX Orin, into autonomous patrol robots. LiDAR and camera input fuse on the edge for real-time environmental perception, anomaly detection, and centralised security monitoring.

The robots work campuses, smart buildings, industrial sites, and public infrastructure — the long, repetitive rounds where human attention fades and machine attention does not.