Embodied Humanoid Robot
Xiaowu Humanoid Robot is a highly versatile embodied AI product designed for industry, commerce, and research.
Features
OBMD-CCUA
Our core algorithmic architecture enables universal deployment across diverse hardware, including humanoid, quadrupedal, and wheeled robots, disrupting the traditional "one software per robot" paradigm. Powered by an atomic skill library, varied robot morphologies seamlessly share core capabilities like grasping, walking, and obstacle avoidance, drastically slashing R&D and maintenance costs across multiple product lines.
OBMC | C-C Co-Decision
Embodied Cerebrum (Decision)Driven by LLMs for language, perception, and long-horizon planning to autonomously decompose complex commands.Embodied Cerebellum (Execution)Driven by WBC for real-time, whole-body coordinated control to precisely execute multi-scenario tasks.
Open Eco | Dev-Friendly
SDK Toolchain Launched: Open for skill invocation, agent configuration, and scenario deployment kits, significantly lowering development barriers.Standardized Dataset Released: Launching the industry's first universal dataset and updating VLA toolchains to drive technology sharing.Open-Source Collaboration: Attracting global developers to optimize algorithms and build an inclusive, efficient embodied AI ecosystem.
Standardized HW/SW Decoupling
Cognitive & Control DecouplingBy integrating the Model Context Protocol (MCP), the Agent acts as a client to issue commands, successfully decoupling high-level cognition from low-level control.Hybrid Control ManagementThe MCP Server provides hybrid orchestration, deeply fusing advanced end-to-end VLA skills while fully compatible with traditional modular control to meet complex industrial needs.
Key Specifications
| Specs | Wheeled-Arm |
| Height (mm) | 1280~1630 |
| Wingspan (mm) | 600 |
| Net Weight(kg) | 88 |
| Color | White + Black |
| Material | Aluminum Alloy + Engineering Plastics |
| DoF | 35 Integrated Joints |
| Robotic Arm | 7 DoF * 2, Max Payload 4.5kg per arm |
| Dexterous Hand | 3kg Payload per hand |
| 6-DoF Force Sensor | Yes |
| Cameras | Head: 1* RGBD / Chassis: 2* RGBD, 1* RGB |
| Audio | MIC Array: 4* Linear MICs / 1* Speaker / 1* Sound Card / 1* 720P RGB Camera |
| Motion Compute | Intel x86 i7-1355U |
| Dev Compute | Model：OrinAGX*1，275TOPS |
| OS | Ubuntu |
| Connectivity | WiFi6，Ethernet，Bluetooth3.3 |
| Charge Time(h) | 2 |
| Battery Life | 4 |
| Specs | Humanoid |
| Height (mm) | 1730 |
| Wingspan (mm) | 600 |
| Net Weight(kg) | 73 |
| Color | White + Black |
| Material | Aluminum Alloy + Engineering Plastics |
| DoF | 42 Integrated Joints |
| Robotic Arm | 7 DoF * 2, Max Payload 4.5kg per arm |
| Dexterous Hand | 3kg Payload per hand |
| 6-DoF Force Sensor | Yes |
| Cameras | 3* Structured Light Depth Cameras (Head, Chest, Waist, Back) |
| Audio | MIC Array: 4* Linear MICs / 1* Speaker / 1* Sound Card / 1* 720P RGB Camera |
| Motion Compute | Intel x86 i7-1355U |
| Dev Compute | Model：OrinAGX*2，550TOPS |
| OS | Ubuntu |
| Connectivity | Full-Body CAN/Ethernet, Bluetooth 3.3 |
| Charge Time(h) | 2 |
| Battery Life | 3.5h |
System Architecture
Task Planning
Cloud/local large models serve as the core "brain," managing multimodal input comprehension (voice and system commands), memory retention, and task orchestration.
MCP Hybrid Management
It fully supports end-to-end VLA skills for tasks like picking and sorting, while remaining compatible with traditional modular methods for handling, inspection, and mobility, flexibly adapting to diverse scenarios.
Toolchains & Simulation
Equipped with a complete data collection, model training, and simulation verification pipeline, it empowers robots to continuously evolve from static execution tools into highly autonomous intelligent agents.
Deployment Process
Process-Based Scheme Orchestration
Ideal for structured environments with highly fixed target objects requiring predictable process execution. Through a workflow spanning process decomposition, input/output interface design, skill code development, and integrated system testing, it achieves final system validation and operational setup. Its core advantages lie in industrial environment reliability and high repetition consistency.
Vision-Language-Action (VLA) Tech Solution
Ideal for dynamic environments with highly variable target objects requiring robust generalization. Through a workflow spanning Agent and VLA configuration, prompt engineering, data collection, model fine-tuning, and testing, it achieves final model validation and operational setup. Its core advantages lie in unstructured environment adaptation and continuous evolution through autonomous learning.
Application Scenarios
In automotive 4S stores, showrooms, and commercial spaces, humanoid robots handle vehicle presentations, customer welcoming, intelligent Q&A, QR-code test-drive bookings, promotional broadcasting, and after-sales guidance. By leveraging voice emotion detection, the system detects negative customer sentiment to trigger automatic handovers to human sales consultants, while supporting real-time multi-model knowledge base switching and seamless CRM integration.
Our humanoid robots operate seamlessly in power distribution rooms, conducting visual meter and status inspections, infrared thermography, partial discharge testing, SF₆ gas/environmental monitoring, and switchgear operation. Empowered by laser SLAM navigation for narrow alleys, the system provides 7×24h unmanned operations, tiered anomaly alerts (emergency/general/reminder), and full-process recording for compliance auditing, while supporting remote teleoperation and multi-robot fleet dispatching.
