Overview
Perception, Decision, and Action — An Integrated Edge AI Robotics Platform
MARS400 PRO is NexCOBOT's edge computing controller built for robotic vision-based grasping applications. Powered by an NVIDIA GPU edge computing platform, it comes preloaded with Isaac ROS Manipulator and a full ROS2 development environment, and can be extended with VLA / VLM models depending on the application. Bundled with the RealSense D435i depth camera, hardware and AI models are delivered as an integrated package — eliminating extra procurement and compatibility concerns, so the system runs vision-based grasping applications right out of the box.
Complete Solution
MARS400 PRO Complete Solution
Edge computing controller + depth camera, sold as a hardware bundle and ready to use out of the box — saving the time and cost of sourcing components and verifying compatibility yourself.
Edge AI Controller
MARS400 PRO
- NVIDIA GPU edge computing platform
- Preloaded with Isaac ROS Manipulator, ROS2
- Supports a variety of VLA / VLM models
- Full ROS2 development environment
RGB-D Depth Sensing Camera
RealSense D435i
- RGB + Depth
- USB plug-and-play
- Provides the image source for FoundationPose pose estimation
- Image source for VLM / VLA
Hardware Bundle, Ready to Use
Physical AI Controller
- Hardware and AI models delivered as an integrated package
- No extra procurement, no compatibility concerns
- Ready to run vision-based grasping applications out of the box (Model training required by the user — NexCOBOT training available)
- Perception, decision, and action — fully integrated
Module Overview
MARS400 PRO Module Overview
Built on isaac_ros_manipulator and integrating multiple VLA / VLM / WBHL models to give users flexible options. Three core modules operate around the MARS400 PRO platform, covering everything from standardized operations to forward-looking bipedal locomotion.
Core Integration | Available
Isaac ROS Manipulator
- Isaac ROS RTDETR
- Isaac ROS FoundationPose
- Isaac ROS cuMotion
Use Case → Pick & place, sorting. For common, simple point-to-point operations, the vision model can run directly after retraining — ideal for production tasks that need fast deployment.
Planned
VLA Models
- Nvidia GR00T
- Physical Intelligence π-0
- VLM
Use Case → Assembly, dual-arm collaboration, packaging. Users can train models for their specific scenarios to extend into more complex or non-standardized tasks.
Gait Control
Whole-Body Humanoid Locomotion <Safetics>
Use Case → Bipedal robot self-balancing. Robot locomotion across enclosed and open-field environments.
System Architecture
MARS400 PRO Software & Hardware Stack
From AI application packages down to the underlying hardware platform, MARS400 PRO integrates a complete, ready-to-run software and hardware stack.
Cross-Brand Robot Control
Cross-Brand Robot Control IntegrationComing Soon
Goal: control robot arms across multiple brands directly or indirectly, so users' existing equipment can adopt MARS400 PRO's AI capabilities.
Roadmap Feature — Not Yet Available Everything in this section is a planned capability under development. Cross-brand robot control is not currently supported on MARS400 PRO and is not part of the shipping product today.
Planned — Not Yet Available
5Target Robot Brands
3Control Methods
Supported Robot Brands
Fanuc
Yaskawa
Kuka
ABB
UR
Robot Control Methods
MARS400 PRO
AI Model
Target Robot
(5 Brands)
›
›
›
›
›
›
›
›
›
Method 1Direct Control
ROS Action / Topic
ros_controller
Method 2Syntax Conversion Control
ROS2 Node (Get Waypoint)
Convert to Robot-Specific Syntax
Method 3GRC Control
ROS2 Node (Get Waypoint)
GRC
