Introduction

Update history
Date Version Author Update content
2026-10-27 v0.1 lxowalle Initial release

MaixSense D640 is a stereo depth camera module from Sipeed. Two global-shutter cameras capture images, and the module calculates depth internally. It sends depth and stereo images over USB for spatial sensing, desktop distance measurement, and robot vision.

MaixSense D640 depth camera

Key features

  • On-device depth processing: Receive depth and stereo images over UVC without calculating stereo depth on the computer.
  • Two baseline options: Choose 50 mm for close-range work or 100 mm for longer distances.
  • Global-shutter stereo capture: Capture matching depth and left/right grayscale images as synchronized frame sets.
  • Image and point cloud viewing: Preview video in a browser, or use the desktop app for distance measurement, 3D viewing, and PLY export.
  • Tools for development: Save synchronized snapshots, record depth data in MCAP, and integrate the camera through the C/C++ SDK or ROS2 driver.

Choose a hardware version

The baseline is the distance between the two cameras. Choose it to match your working distance:

Front and rear views of the short- and long-baseline versions
Version Measurement range Typical use
Short baseline, 50 mm 0.25–2.0 m Robot wrist mounting, close-range sensing, desktop measurement
Long baseline, 100 mm 0.5–4.0 m Robot head mounting and medium-range sensing on cleaning, lawn-mowing, or service robots

Example applications:

Robotic hand application Food delivery robot application Robot vision application Robot vacuum application

Specifications

Item Specification
Processor Dual-core processor with integrated 2.4 TOPS NPU
Image sensors Two global-shutter sensors, 640 × 480 pixels each
Optical filter Dual-band 650 nm / 850 nm bandpass filter
Depth resolution 640 × 384 pixels
Depth frame rate 12 fps
Infrared frame rate 30 fps
Baseline Short: 50 mm. Long: 100 mm
Measurement range Short: 0.25–2.0 m. Long: 0.5–4.0 m
Field of view 57° horizontal × 43° vertical, 70° diagonal
Data interface USB 2.0, USB Type-C connector
Power 5 V DC, supply capable of at least 0.5 A
Dimensions Short: 65 × 22 × 16.04 mm
Long: 112 × 22 × 16.04 mm
Weight Short: 35 g
Long: 60 g

Short baseline:

Short-baseline version dimensions

Long baseline:

Long-baseline version dimensions

Software and guides

Start with Quick Start to connect the camera. Use MaixSense Tool to view images, measure distances, or record depth data. Open the D640 online preview to view one video stream in a browser. Use the SDK for programming.

Tool What it does
Online preview View and record a pseudocolor or infrared video stream. See Quick Start
MaixSense Tool Desktop viewing, distance measurement, point clouds, snapshots, MCAP recording, and device updates. See Desktop App Guide for Windows, Linux, and macOS availability
SDK C/C++ capture, synchronized frame sets, filtering, and point clouds. See SDK Guide
ROS2 View images and point clouds, save data, and record sensor messages. See ROS2 Guide
Agent Capture images, save files, and check results through a conversation. See Agent Guide

MaixSense Tool main window

Image outputs

Use raw depth for measurement and point clouds. Use the device's pseudocolor stream for a convenient view of relative distance.

Output Format Resolution Frame rate Use
Raw depth Z16/Y16 640 × 384 12 fps Millimetre distance values for measurement and point clouds
Pseudocolor preview MJPEG 640 × 384 12 fps Color represents distance. This stream cannot provide metric depth
Left infrared MJPEG 640 × 480 30 fps Left camera grayscale image
Right infrared MJPEG 640 × 480 30 fps Right camera grayscale image

Raw 16-bit depth often looks dark when displayed directly as grayscale. This is expected: the pixel values represent millimetres rather than normal image brightness.

Raw depth displayed as grayscale

Device pseudocolor preview

The camera's raw depth and device-generated pseudocolor streams cannot be enabled together. The desktop app and ROS2 can instead colorize raw depth on the computer while retaining the original depth values.

Next steps