Hiwonder NexArm Open-Source ROS Robotic Arm VLA Embodied AI Kit with NVIDIA/Raspberry Pi & ESP32 Dual-Controller Design, Multimodal Large AI Models & OpenClaw (Without Raspberry Pi Kit)

• 【Embodied AI ROS 2 Robotic Arm】ROS 2-powered & Jetson/Raspberry Pi 5 compatible. Packed with a 3D depth camera, AI voice box, and 65 kg·cm magnetic encoder servos to effortlessly deliver motion-sensing interaction, voice control, and intelligent grasping.
• 【Large AI Models & OpenClaw Agent】Supports online/local multimodal AI model deployment. Deep OpenClaw Agent integration enables remote voice/text interaction, autonomous task decomposition, intelligent decision-making, and complex task execution.
• 【Integrated Algorithm Framework】Integrates YOLO26, OpenCV, MediaPipe, Gmapping, inverse kinematics, and other algorithms and tools, providing a highly configurable and extensible development environment.
• 【Visual PC Software】Features professional PC software supporting one-click scene switching and dual-stream real-time video feedback. Easily customize scenes and hardware parameters for smooth, efficient operation.
• 【Out-of-the-Box Ready & Full-Scale Curriculum】Pre-loaded with Ubuntu 22.04 alongside a complete dev stack of Python, TensorFlow, PyTorch, and ROS. Includes an end-to-end curriculum spanning computer vision, deep learning, NLP, inverse kinematics, large AI models, and VLA imitation learning.

Note: Aside from the controller versions available on our website, if you are looking for other Raspberry Pi or Jetson controller versions, please contact us directly to make a purchase.

Product Description:

Hiwonder's NexArm ROS Robotic Arm is an all-in-one experimental platform purpose-built for advanced ROS education and embodied AI research. Featuring an advanced modular architecture, it natively fuses multimodal large models, OpenClaw agents, ACT, VLA, machine vision, and inverse kinematics into a complete validation loop spanning foundational concepts to high-level embodied algorithms.

Its NVIDIA/Raspberry Pi + ESP32 dual-controller setup powers a high-performance hardware suite consisting of magnetic encoder bus servos, a parallel-rail gripper, a 3D depth camera, and an AI voice module. Deep integration of multimodal AI and OpenClaw agents drives precise 3D depth perception and multi-dimensional sorting. It fully satisfies full-stack development demands, spanning environmental sensing to real-world physical manipulation.

1) Large AI Model & OpenClaw

The expandable NexArm ROS robotic arm seamlessly integrates with advanced peripherals, including electric sliding rails and the suspended chassis. The modular metal sandbox baseplate enables highly efficient replication of complex physical interaction scenarios like smart warehousing.

2) Versatile Controller Compatibility

The NexArm ROS robotic arm controller assembly supports Raspberry Pi and high-performance Jetson controllers. Equipped with a multi-functional expansion board and rich ports, it seamlessly accelerates development and hardware expansion.

3) 3D Spatial Recognition & Grasping

An integrated 3D structured light depth camera fuses RGB and depth data, utilizing advanced vision algorithms to power object tracking, detection, and facial recognition for precise 3D spatial perception and grasping.

4) Inverse Kinematics & Eye-Hand Coordination

Powered by an in-house, high-performance inverse kinematics algorithm, NexArm reaches any coordinate within its workspace, driving advanced AI projects like grasping, sorting, and transporting.

5) Dual-Output-Shaft Magnetic Encoder Bus Servo

A dual-shaft design with the main shaft for power output and the secondary shaft for support, delivering greater operating stability.

6) Parallel-Rail Gripper

Silicone gripper pads and stainless-steel transmission components ensure a secure grip and improved durability.

7) Aerospace-Grade Industrial Bearing Base

An industrial bearing base supports 270° joint rotation and positioning, while the aerospace-grade structure reduces Z-axis wobble.

8) OLED Display Design

Equipped with an OLED screen that displays real-time data of this open-source robotic arm.

1. Multimodal Large Al Models for Embodied Intelligence

1) Three Large Al Models Deployed

① Large Language Model - Acts as the brain of the robot, enabling it to understand commands and deliver intelligent responses.

② Large Speech Model - Gives the robot ears and a voice,enabling voice recognition and natural speech interaction.

③ Vision Language Model - Serves as the eyes of the robot, enabling accurate object recognition and detailed visual understanding.

2) Built-In OpenClaw Al Agent

① Cross-Platform Collaboration

Breaking single-terminal limits, the system integrates with Telegram and supports remote PC.app, and local voice multimodal interactions.

② Task Decomposition & Scheduling

High-level commands decompose into atomic actions like navigation, recognition, and reporting, invoking low-level interfaces for an end-to-end closed loop.

③ Memory & Proactive Feedback

Retaining past commands and environmental states, the robot autonomously handles exceptions, proactively reporting and confirming execution strategies for a highly intuitive interaction experience.

2. Multimodal AI Model Creative Applications

1) Smart Waste Sorting

The NexArm ROS robotic arm recognizes diverse sorting cards within its field of view and classifies them on command.

2) Visual Object Tracking

Powered by large vision models, NexArm ROS Instantly locks onto and tracks target objects dynamically.

3) Visual Scene Understanding

NexArm ROS percelves environmental features and object attributes within its workspace, delivering natural voice feedback.

4) Smart Home Assistant

Transmitting real-time camera feeds to large vision models, NexArm ROS generates coordinated multimodal responses In speech and actlon.

5) Intelligent Transport

Dynamically adjusting its pose in real time, NexArm ROS robotic arm precisely grasps and transports target objects based on Interactive Instructlons.

3. Large AI Model × OpenClaw Automated Intelligent Sorting

1) Large Al Model x OpenClaw x Intelligent Information Aggregation

NexArm ROS robotic arm captures visual metrics, analyzes real-time data, and automatically aggregates environmental detalls Into Intuitive decislon reports-seamlessly closing the loop fromon-siteperception to data delivery.

2) Large Al Model x OpenClaw x Proactive Response

When encountering operational anomalies or vague commands, NexArm ROS actively audits the scene, proposes optimal alternatives, and requests authorization to ensure uninterrupted workflows.

3) Large Al Model x OpenClaw x Autonomous Task Deconstruction

NexArm ROS deconstructs remote commands into precise action primitive sequences, directly translating vague semantics into deterministic OpenClaw execution.

4) Large Al Model x OpenClaw x 3D Grasping & Scene Understanding

Large-model-driven human cognition empowers NexArm ROs to deeply comprehend unstructured 3D environments, flexibly mapping semantic intent to precise 3D grasping and sorting actions to close the embodied Al loop.

5) Large Al Model x OpenClaw x Dynamic Low-Level Invocation

Evaluating environment and object states in real time, NexArm ROS autonomously invokes high-performance IK algorithms and low-level execution routines for grasping and transporting to maximize complex task efficiency.

6) Large Al Model x OpenClaw x Spatial Vector Memory

Leveraging interaction-driven environmental memory. NexArm ROS correlates historical targets with current scenes to automatically reuse learned policies, achieving an efficient cognitive-action loop.

4. Versatile Scenario Expansion

Supports mobile suspended chassis and electric sliding rails to precisely simulate smart warehousing, logistics sorting, and automated production lines for rapid Embodied AI algorithm validation.

1) Standard Sorting Sandbox

Equipped with sorting trays and color blocks to handle complex tasks like visual color recognition, multi-colored blocks classification, transporting, and intelligent sorting.

2) Advanced Sorting Sandbox

Adds physical props like miniature sorting bins to expand capabilities into automated waste classification, sorting and transporting of multi-colored blocks, and intelligent sorting control.

3) Sliding Rail Warehouse Sandbox

Integrates a high-precision electric linear sliding rail and multi-tier shelving. It supports multi-workstation transport, automated inventory sorting, and inbound control to fully replicate industrial warehousing and logistics workflows.

4) Suspended Mobile Chassis

NexArm ROS is integrated with a suspended tank or suspended omnidirectional Mecanum wheel chassis. Beyond supporting remote mobile control via wireless controller or app, the setup drives efficient mobile tracking and autonomous line following.

5. ROS 2 Core Features

1) 3D Depth Camera & 3D Depth Perception

Equipped with a 3D structured light depth camera, the NexArm ROs robotic arm precisely calculates real-time 3D

spatial data, including distance, shape, and volume. Native compatibility with popular algorithm frameworks like

OpenCV, KCF,YOLO26, and MediaPipe delivers a rich,highly efficient machine vision development.

3D Depth Point Cloud Images

Distance Measurement

Height Measurement

Volume Measurement

KCF Object Tracking

3D Object Detection

YOLO26 Object Detection

OpenCV Color Recognition

OpenCV Tag Recognition

MediaPipe 3D Face Detection MediaPipe Gesture Recognition

MediaPipe Pose Estimation

2) Deep Vision & Hand-Eye Coordination

By fusing RGB and depth data, NexArm ROS accurately senses color and point cloud information, enriching the geometric representation of spatial environments. Powered by advanced inverse kinematics, NexArm ROS seamlessly executes complex Al tasks, including arbitrary 3D grasping, sorting, and transporting.

Arbitrary 3D Grasping

Voice-Controlled

Color Sorting & Tag Sorting

Smart Palletizing

Item Sorting

Waste Sorting

Face Tracking

Motion-Sensing Control

3) Advanced Inverse Kinematics Algorithms

With an advanced In-house IK algorlthm, NexArm ROS's end-effector seamlessly accesses any coordinate within its workspace. Easily customize adaptive grasping, linear motion, and advanced path planning via programming for a pure, distractlon-free development environment. (Official IK function APls included)

Hand-Eye Coordination

Adaptive Joint Adjustment

Multi-Dimensional

Path Planning

Motion Control

Linear Motion

PC Software Coordinate Control

Code-Based Coordinate Control

Provided Function APIs

4) Gazebo Simulation & Movelt 2 Motion Planning

Built on the ROs 2 framework, NexArm ROs provides robust Gazebo and Movelt simulation support for virtual motion control testing and algorithm validation, bypassing physical limitations to dramatically lower development overhead.

URDF Kinematic Design

Moveit Simulation Control

Movelt Motion Planning

Moveit Scene Design

Collision Detection

Virtual Linkage

Sim-To-Real Linkage

Complex Scene Building

6. Professional Host PC Software

Designed specifically for NexArm, this all-in-one platform handles coordinate control and motion trajectory planning. Supports precise PID and acceleration tuning,enables real-time monitoring of arm operation, and integrates quick access to peripheral coordination and Al features, balancing beginner-friendly operation with professional development needs.

1) Multiple Operation Modes with Integrated Control

① Action Group Mode

Edit joint poses frame by frame with intuitive sliders or numeric input, making it easy to create smooth motion sequences

② Coordinate Control Mode

Powered by inverse kinematics, this mode allows direct input of 3D coordinates and pose parameters to drive the end effector precisely to the target position.

③ 3D Control Mode

A 3D visualization interface mirrors the real arm in real time, making spatial pose clear at a glance and motion planning more intuitive.

④ Robotic Arm System Settings

Customize app & wireless controller parameters, fine-tune servo settings. speed, and acceleration to match different control preferences & applications.

2) Precision Joint Tuning with System-Wide Monitoring

① Multiple Safety Protections

Professional PC software allows customizable safety parameters, including overload limits and torque thresholds, helping ensure stable and reliable operation.

② PID Tuning Support

Servo PID parameters can be read and adjusted flexibly to improve motion stability and response speed for better overall performance.

③ Adjustable Speed and Torque

Servo speed. torque, and other parameters can be configured as needed to maximize performance for high-precision control tasks.

④ Real-Time Parameter Feedback

Voltage, position, angle, speed, torque thresholds, and other parameters are sampled at high frequency. Feedback in real time, and displayed visually throughout operation.