Isaac ROS-Compatible Robotics AI SDK

Deploy at Scale
with DEEPX

One-Line Migration from Isaac ROS

Swap a supported Isaac ROS package for its DX-Newton equivalent. Keep the same ROS 2 topics, messages, launch structure, and downstream nodes.

Production-Ready Edge Efficiency

 Bring your ONNX or supported PyTorch models into DXNN and move inference to a DEEPX NPU in hours.

Replace

isaac_ros_yolov8

with

, then

press Enter.

perception.launch.py

1 line to edit

9

ComposableNode(

10

name='yolov8_node',

11

package='

isaac_ros_yolov8

',

12

plugin='Yolov8DecoderNode',

13

remappings=[('image','/img_raw')],

14

),

Replace isaac_ros_yolov8 with newton_yolov8.

*Third-party names are trademarks of their respective owners and are referenced here to describe compatibility only. Package names shown for layout review — final list pending R&D confirmation.

Reference GPU Platform VS. DEEPX NPU

Same Perception Task.
A Fraction of the Power.

Across the benchmark configurations shown below, DX-Newton runs supported robotic-perception workloads on the DEEPX NPU with comparable task performance and substantially lower accelerator power than the referenced GPU platform.

Reference GPU + Isaac_ros

Visual SLAM

Power Consumption

~47

W

Operating Temperature

~64

°C

Cooling Configuration

DEEPX NPU + Newton_ros

Visual SLAM

Power Consumption

~2.2

W

Operating Temperature

~40

°C

Cooling Configuration

Reference GPU + Isaac_ros

Stereo Depth

Power Consumption

~60

W

Operating Temperature

~76

°C

Cooling Configuration

DEEPX NPU + Newton_ros

Stereo Depth

Power Consumption

~3.6

W

Operating Temperature

~46

°C

Cooling Configuration

Reference GPU + Isaac_ros

Pose Estimation

Power Consumption

~59

W

Operating Temperature

~74

°C

Cooling Configuration

DEEPX NPU + Newton_ros

Pose Estimation

Power Consumption

~2.9

W

Operating Temperature

~43

°C

Cooling Configuration

Reference GPU + Isaac_ros

AprilTag Detection

Power Consumption

~43

W

Operating Temperature

~58

°C

Cooling Configuration

DEEPX NPU + Newton_ros

AprilTag Detection

Power Consumption

~1.7

W

Operating Temperature

~37

°C

Cooling Configuration

Reference GPU + Isaac_ros

Object Detection

Power Consumption

~50

W

Operating Temperature

~66

°C

Cooling Configuration

DEEPX NPU + Newton_ros

Object Detection

Power Consumption

~2.6

W

Operating Temperature

~42

°C

Cooling Configuration

Reference GPU + Isaac_ros

Semantic Segmentation

Power Consumption

~54

W

Operating Temperature

~70

°C

Cooling Configuration

DEEPX NPU + Newton_ros

Semantic Segmentation

Power Consumption

~3.2

W

Operating Temperature

~44

°C

Cooling Configuration

Visual SLAM

Reference GPU + Isaac_ros

Visual SLAM

Power Consumption

~47

W

Operating Temperature

~64

°C

Cooling Configuration

DEEPX NPU + Newton_ros

Visual SLAM

Power Consumption

~2.2

W

Operating Temperature

~40

°C

Cooling Configuration

Stereo Depth

Reference GPU + Isaac_ros

Stereo Depth

Power Consumption

~60

W

Operating Temperature

~76

°C

Cooling Configuration

DEEPX NPU + Newton_ros

Stereo Depth

Power Consumption

~3.6

W

Operating Temperature

~46

°C

Cooling Configuration

Pose Estimation

Reference GPU + Isaac_ros

Pose Estimation

Power Consumption

~59

W

Operating Temperature

~74

°C

Cooling Configuration

DEEPX NPU + Newton_ros

Pose Estimation

Power Consumption

~2.9

W

Operating Temperature

~43

°C

Cooling Configuration

AprilTag Detection

Reference GPU + Isaac_ros

AprilTag Detection

Power Consumption

~43

W

Operating Temperature

~58

°C

Cooling Configuration

DEEPX NPU + Newton_ros

AprilTag Detection

Power Consumption

~1.7

W

Operating Temperature

~37

°C

Cooling Configuration

Object Detection

Reference GPU + Isaac_ros

Object Detection

Power Consumption

~50

W

Operating Temperature

~66

°C

Cooling Configuration

DEEPX NPU + Newton_ros

Object Detection

Power Consumption

~2.6

W

Operating Temperature

~42

°C

Cooling Configuration

Semantic Segmentation

Reference GPU + Isaac_ros

Semantic Segmentation

Power Consumption

~54

W

Operating Temperature

~70

°C

Cooling Configuration

DEEPX NPU + Newton_ros

Semantic Segmentation

Power Consumption

~3.2

W

Operating Temperature

~44

°C

Cooling Configuration

*Results are provided for directional comparison only and are not based on a side-by-side hardware test. Actual results may vary.
*GPU: Power (W), Temp (°C) (Published Reference) | DEEPX: Power (W), Temp (°C) (Simulated Estimate)
*GPU figures are based on publicly available specifications and benchmark data.
*DEEPX figures are simulation-based engineering estimates.

Full Stack Architecture

Replace the Engine, Not the Robot.

Keep your ROS 2 application graph. DX-Newton replaces the underlying AI acceleration stack across algorithms, runtime, memory, and compute backends.

Application & ROS 2 nodes

Perceptiondetection · segmentation · depth · pose · face · hand · OCR · audio

Navigationvisual SLAM · 3D mapping · localization · freespace

Manipulation & controlmotion planning · robot segmenter · action policy

Processingimage proc · stereo proc · pointcloud · encode · codec

Toolingvisualization · benchmark · contract diff

DX-Newton Integration & Abstraction

For supported packages, update the package reference in your launch file while preserving the ROS 2 node graph, topics, message types, parameters, and downstream interfaces.

Robotics AI & Vision Libraries

Stereo Visual Odometry

Visual place recognition

AI Stereo Depth + SGM

Monocular Depth Estimation

Detection & segmentation DNN

Open-Vocabulary Detection

2D / 6DoF Pose Estimation

AprilTag Detection

3D reconstruction

Image & codec ops

Inverse Kinematics & Collision-Aware Planning

Pre-Optimized Models from the DEEPX Model Zoo

Your Model → Compile with DX-COM → Optimize & Profile

Runtime, Transport & Memory

Inference RuntimeHardware-abstracted inference backend

Model ProfilesPerformance and accuracy validation

Transport & MemoryOptimized, zero-copy-capable ROS 2 transport

Hardware Abstraction Layer

Backend selection, execution policies, and accuracy validation. Switch supported compute backends without redesigning the ROS 2 application graph.

Supported Compute & Media Hardware

DEEPX NPU: DX-M1

DEEPX NPU: DX-M2 Planned

2D Vision Engine

Optional GPU Compute

Hardware Codec

CPU Fallback

Stack Ownership

Existing ROS 2 Application

DX-Newton Integration & Abstraction

DX-Newton Software Stack

Compute & Media Hardware

Across the Stack

DNN PipelinePre-process · Infer · Post-process

Supported ROS 2 ReleasesJazzy · Humble

MiddlewareRMW · DDS

Verification

Robotics Solutions

Built for Robots That Have to Ship.

Across humanoids, industrial robots, mobile robots, service robots, and drones, power, thermal design, and compute cost determine whether an AI prototype can scale into a production-ready system.

Humanoids

Fanless operation frees the head and torso volume that active
cooling used to occupy.

Industrial Robots & Arms

Run multi-camera inspection, pose estimation, and vision-guided
automation within tighter power, thermal, and cost budgets.

AMR & AGV

Lower AI power leaves more battery capacity for motion, payload,
and longer operation between charging cycles.

Service & Delivery Robots

Quiet, low-heat AI compute helps robots operate naturally in
lobbies, hospitals, stores, and other human-centered spaces.

Drones & UAVs

Every watt saved and gram of cooling hardware removed can
translate into more payload capacity or longer mission time.

Silicon Platforms

One DX-Newton Interface.
From Perception to Multimodal AI.

DX-M1 accelerates today’s robotic-perception workloads within a single-digit-watt NPU power envelope. DX-M2 is designed to extend the DX-Newton interface to multimodal and generative AI, helping teams expand capability without redesigning the ROS 2 application stack.

DX-M1

Available

Efficient Perception at the Edge

Accelerate object detection, segmentation, pose estimation, tracking, and re-identification on the DX-M1 NPU through DX-Newton.

DX-M2

Upcoming

Multimodal AI at the Edge

Designed to bring vision-language understanding, instruction processing, and vision-language-action workloads to power-constrained robots through DX-Newton.

Get In Touch

Talk to the DX-Newton Team.

Whether you are evaluating model compatibility, planning a ROS 2 deployment, or exploring a production partnership, tell us about your project and we will connect you with the right DEEPX team.

Technical Support

Model compatibility and porting, supported operators, ROS 2
integration, and performance benchmarking.

Business & Partnerships

Evaluation hardware, volume pricing, production planning, and
technology partnerships.

First name*

Last name*

Company or organization*

Work email*

How can we help ?*

Which robotics software stack are you currently using ?*

Phone number (optional)

Are you evaluating a migration from Isaac ROS ?*YesNoNot sure yet

Tell us about your project.*

Please include the robot type, AI workloads, current compute platform, development stage, target schedule, and expected volume.

By submitting, you agree to the Privacy Policy and agree to receive occasional emails.