Machine Vision Module — Intelligent Perception，Unleash Infinite Design Flexibility，Native Inheritance，Significantly save development time

From Pixel Logic to Physical Intelligence

High-Fidelity Data, Driving the Evolution of Embodied Intelligence

Hardware Sync
Eliminates millisecond-level latency between vision and action, providing precise causal training data.
Physical Truth
Sub-millimeter depth accuracy, perfectly reproducing the 3D logic of the physical world.
Accelerated GeneralizationHigh-quality multimodal data, shortening the Sim-to-Real gap for AI.

Redefining the Benchmark for Spatial Perception

Core Components of Embodied Intelligence: Perception + Decision

Edge-Side Powerful Compute, Unleashing the Core Capability of the Main Processor

Integrated High-Performance SoC:：
The module integrates a high-compute SoC, achieving "perception as computation". By running high-speed, high-precision VIO, SLAM/CSLAM, and real-time loop closure detection，independently on the edge, complex spatial computations are completed within the module.

Independent Compute Engine (Compute Offload):
The module directly outputs high-value pose and localization results,significantly reducing the compute overhead of the main controller (host).This allows the robot's "brain" to focus more on high-level algorithms like path planning and task decision-making, markedly improving the overall system response speed.

Multi-Mode Assistance:

Supports high-speed QR code, spatial anchor point, and cloud-edge collaborative modes, ensuring absolute localization stability in various complex environments.

Deep Multi-Sensor Synergy: Achieving Accurate Mapping from the Hardware Level Up

Achieves hardware-level sampling synchronization between the image sensor and the IMU inertial unit. Through deep alignment of visual and inertial data, we directly solve the issues of latency and error in spatial motion data at the hardware level. This "natively aligned" data stream not only provides the robot with the most accurate spatial motion ground truth but also significantly improves the response speed (fast), static accuracy (accurate), and dynamic stability (stable) of the SLAM localization algorithm.

Global Shutter, Faithfully Reproducing the Physical Environment

Designed for high dynamic data capture. Optimized for dynamic environments, suitable for rapidly changing viewpoints, such as during robot joint axis rotation. Higher sensitivity ensures image quality in complex lighting conditions, reducing the "rolling shutter effect". Every frame is a deterministic physical ground truth from the real environment.

Built-in CNN Acceleration Engine: On-Edge Real-time Semantic Perception

Integrates a dedicated neural network hardware acceleration unit, supporting real-time execution of object detection, 6DoF pose estimation, and semantic segmentation on the module for localized, real-time AI inference. Combined with VSLAM and depth information, it enables advanced intelligent functions such as 3D gesture recognition and tracking, somatosensory recognition, 3D face anti-spoofing recognition, object recognition and spatial semantic segmentation, scene understanding, and more. Perfectly compatible with OpenCV & Open VINO, allows easy deployment of models from open-source communities (e.g., Open Model Zoo) or user-trained models – plug-and-play, significantly shortening the R&D cycle.

Efficient Integration and Expansion

Ready-to-Use, Flexible Expansion

Intelligent video compression

Hardware-based H.265 video compression significantly saves valuable transmission bandwidth and reduces the overall system load.
Plug and Play

Localized processing capabilities significantly lower the development and integration threshold.
Enrich SDK

Supports Windows/Linux/Android, and provides wrappers for ROS/ROS2, Unity, C/C++, Python, Java JNI, etc. API, seamlessly connect to various development frameworks.
Highly scalable

The USB Type-C is used to connect to the main controller, and sensors such as lidar can be connected through the expansion interface. It supports multi-machine cascading to meet complex system requirements.
Global Certification

Passed CE, FCC, RoHS, CB, FDA Class I laser safety certifications, and is accepted in the global market.

Empowering Countless Industries, Leading the Future

SeerSense® SeerSense® DS80-4 is a shining example of Xvisio's spatial perception technological prowess. The core technology matrix behind it is not only the "smart eyes" enabling humanoid robots to achieve autonomous navigation, dexterous manipulation, and intelligent interaction but also holds broader application potential.

Smart Manufacturing

Humanoid robot/AMR navigation, industrial inspection, and automated sorting.
Daily life

Smart access control and personnel behavior analysis.
Medical

Surgical navigation, rehabilitation assistance, and telemedicine perception.
Logistics and Retail

Volume measurement, shelf analysis, and unmanned delivery.
AR/VR & the Metaverse

High-precision positioning tracking and immersive interaction.
Office and entertainment

Virtual shooting, somatosensory games, virtual meetings.
Smart Home

Service robots and intelligent monitoring.

、

Product Composition

SeerSense® S80-4 Module
SDK
Tool Chain

Specifications

Item	Specification	Notes
Processor	High-performance SOC
Camera
-Quad Fisheye Cameras	1280x800/640x400,50fps DFOV150° /HFOV130° /VFOV74°	Monochrome，Global Shutter 80mm Baseline
-RGB Camera	12MP(4096*3072),30FPS DFOV157.8° /HFOV125.7° /VFOV 98.6°（±3°）	Sensor：VGA/720P/1080P@30fps
IMU	9 Axis,1000Hz	1000Hz for SLAM Fusion
CNN Engine	Dual Engine	Parallel Processing
Binocular Depth Engine	VGA(720P)SGBM	50/60fps,6.5m
SLAM Engine	Avg. RMSE≤0.25%	Robust in complex environments
Power Consumption	Avg. 2.8w	Full load
Power	5V/3A
Connectivity	USB Type-C Debug Port	Camera data, depth, Al image, 6DoF Output UART,Sync Pin
Weight \| Dimensions	93g \| 1192324	LDH(mm)
SDK	VIO,CSLAM,3D Reconstruction, Depth Estimation,Dbject Recognition, Gesture Recognition, Plane Detection,RGBD
Tool Chain	Open VINO,Open CV,Xvsdk-viewer Demo Tool