News Details

Abstract

As one of the earliest embodiments of embodied intelligence to achieve practical industrial deployment, wheeled humanoid robots combine the efficient mobility of AGVs/AMRs with the dexterous manipulation capabilities of humanoid robotic arms. This hybrid architecture is particularly suited to manufacturing and warehousing logistics environments characterized by partially unstructured tasks. Based on industry technology developments, this article examines wheeled humanoid robots from a purely technical perspective, covering their definition, evolution pathway, three-layer technical architecture, key component principles, motion control models, and practical deployment considerations. It provides a detailed analysis of technical mechanisms and future development directions for professionals engaged in R&D, system integration, and industrial operations.

1. Technical Definition and Structural Architecture of Wheeled Humanoid Robots

1.1 Standard Technical Definition

A wheeled humanoid robot is a composite robotic system that combines a humanoid upper-body manipulation structure with a wheeled mobile chassis. Fundamentally, it represents the deep mechatronic integration of a mobile robot platform and a humanoid manipulation system. Its primary design objective is to achieve integrated mobility and manipulation while maintaining high movement efficiency and operational stability.

According to current engineering configurations, mainstream products can be divided into two categories:

• Wheeled chassis with a humanoid upper-body structure: The upper body is rigidly mounted to the chassis without a lifting mechanism, resulting in a fixed center of gravity. This configuration is suitable for flat-floor warehouses and production-line material handling scenarios.

• Wheeled chassis with a lifting platform structure: An integrated linear lifting module enables adjustable working heights, making it suitable for multi-level storage racks and production lines requiring interaction across different workstation heights.

1.2 Core Advantages from an Engineering Perspective

Compared with bipedal humanoid robots, conventional AGVs, and stationary robotic arms, wheeled humanoid robots offer three major engineering advantages in manufacturing and logistics environments.

Dynamic Stability

Benefiting from the static support structure of wheeled chassis systems, wheeled humanoid robots do not require complex dynamic balancing algorithms. This results in lower steady-state errors and greater suitability for continuous industrial operations.

Energy Efficiency and Endurance

Wheel-based locomotion primarily relies on rolling friction. Under equivalent payload conditions, energy consumption can be reduced by more than 50% compared with bipedal walking systems, enabling longer continuous operation cycles.

Hardware Simplicity

The elimination of high-load leg joints and associated hydraulic or servo drive systems significantly reduces overall mechanical complexity. The total number of motion degrees of freedom is typically reduced by approximately 40%, leading to lower failure rates and reduced hardware costs.

1.3 Technology Evolution Path

From the perspective of industrial robotics development, wheeled humanoid robots follow a progressive evolution path from isolated operation to fully integrated mobility and manipulation.

Fixed Robotic Arm Stage

Only fixed-point manipulation capabilities are available, with operational reach limited by the robot arm's working envelope.

AGV/AMR Autonomous Mobility Stage

Autonomous navigation and material transportation capabilities are achieved, but without active manipulation functions.

Composite Robot Stage

Mobile platforms and robotic arms are integrated, yet mobility and manipulation remain independently controlled, resulting in limited coordination.

Wheeled Humanoid Robot Stage

Mechanical systems, algorithms, and perception technologies are deeply integrated, enabling robots to perform grasping, sorting, inspection, and other tasks while moving.

At present, wheeled humanoid robots remain in the growth phase of technological development. Leading manufacturers have already achieved production volumes exceeding thousands of units, while future technological iterations focus primarily on multi-robot collaboration, dynamic force control, and model lightweighting.

2. Overall Technical System: Brain, Cerebellum, and Body Architecture

The complete technical architecture of a wheeled humanoid robot can be compared to a biological organism and divided into three interconnected layers: environmental perception and intelligent decision-making (brain), motion control and coordination (cerebellum), and core components and actuators (body). The performance of the entire system depends on the tight coupling among these three layers.

2.1 Brain: Environmental Perception and Intelligent Decision-Making

The perception and decision-making system serves as the computational and logical center of the robot, responsible for environmental modeling, instruction interpretation, and task planning. It is the core carrier of embodied intelligence.

2.1.1 Multimodal Foundation Model Technology Stack

The mainstream technical route currently adopts a dual-model architecture consisting of a Vision-Language Model (VLM) and a Vision-Language-Action Model (VLA).

The VLM enables bidirectional mapping between visual information and natural language, supporting object recognition, scene understanding, and instruction interpretation. In logistics sorting applications, it can accurately understand commands such as "sort defective containers" or "pick up the tote bin on the left," achieving recognition accuracy exceeding 95%.

The VLA establishes an end-to-end perception-decision-action pipeline, directly converting visual and language inputs into executable action sequences. Industrial applications generally require inference latency below 50 ms per frame.

Future development directions include model compression for edge deployment, cross-domain transfer learning, and deeper integration between VLM and VLA frameworks to eliminate intermediate interpretation stages.

2.1.2 SLAM Technology

SLAM forms the foundation of autonomous navigation for wheeled humanoid robots. Current systems have evolved from single-sensor laser SLAM to multi-sensor fusion SLAM architectures.

Its core function is to estimate the robot's position in unknown environments while simultaneously constructing environmental maps in real time.

Typical sensor configurations include:

• 2D/3D LiDAR

• Depth cameras

• Inertial Measurement Units (IMUs)

Industrial warehousing applications typically require absolute positioning accuracy within ±1 cm and map update frequencies exceeding 10 Hz in dynamic environments.

An emerging trend is the integration of digital twin technologies with SLAM systems, enabling path planning and obstacle prediction within virtual environments before deployment.

2.1.3 3D Vision Perception and Semantic Recognition

Using depth cameras to generate three-dimensional point clouds and AI-based segmentation algorithms, wheeled humanoid robots can classify obstacles and assess operational risks.

Capabilities include:

• Distinguishing navigable areas from static and dynamic obstacles

• Identifying personnel, storage containers, cables, racks, and equipment

• Assigning risk levels and executing differentiated avoidance strategies

Key technical challenges remain robustness under strong lighting, reflective surfaces, and partially occluded industrial environments.

2.2 Cerebellum: Motion Control and Coordination System

The motion control system manages chassis mobility, robotic arm movement, and overall center-of-gravity coordination. It is a critical subsystem for ensuring operational precision and stability.

Motion control companies such as Yikong Intelligent have developed mature solutions in multi-axis coordination, dynamic force control, and system calibration, significantly shortening the transition from research and development to industrial deployment.

2.2.1 Wheeled Chassis Motion Control Models

Mainstream chassis drive architectures include differential drive and omnidirectional drive (Mecanum wheel) systems.

Differential Drive

This configuration features a simple structure and high payload capacity, making it suitable for heavy-duty warehousing applications. Kinematic calculations are based on wheel speed differentials and chassis geometric parameters.

Omnidirectional Drive

By utilizing Mecanum wheels, the robot can perform lateral movement and zero-radius rotation. Motion calculations rely on matrix transformations between wheel velocities and chassis velocities.

Current optimization priorities include:

• Response times below 100 ms for step commands

• Ground-adaptive control for floor joints and slight slopes

• Coordinated control during simultaneous movement and manipulation

2.2.2 Upper-Body Manipulation Control

Control of multi-degree-of-freedom robotic arms and end effectors typically combines position control with force-position hybrid control.

Trajectory planning generally employs fifth-order polynomial interpolation to ensure smooth acceleration and deceleration while minimizing mechanical impact.

Force-position hybrid control simultaneously constrains end-effector position and contact force, making it suitable for fragile products and precision assembly tasks.

When large-arm movements occur, coordinated torso rotation helps maintain system stability by adjusting the robot's center of gravity.

2.2.3 Mainstream Control Architectures

Most industrial systems employ hybrid control frameworks combining model-based control and reinforcement learning.

Model-based control ensures trajectory accuracy and repeatability, while reinforcement learning enhances adaptability to changing environments. Together, they provide both precision and flexibility.

2.3 Body: Core Components and Actuation Systems

Core components and actuators determine payload capacity, accuracy, service life, and environmental protection performance.

2.3.1 Drive and Transmission Components

Servo Motors

Chassis systems typically utilize low-voltage servo motors or hub motors, requiring high torque output, strong overload capability, and protection ratings above IP54.

Upper-body joints often employ coreless motors or frameless torque motors to achieve high power density and lightweight construction.

Reducers

Heavy-load joints such as shoulders, waists, and chassis steering mechanisms commonly use planetary gear reducers or worm gear reducers.

Precision joints, including wrists and dexterous hands, generally employ harmonic reducers with minimal backlash and transmission efficiencies exceeding 85%.

2.3.2 End Effectors

Common end-effector configurations in logistics applications include:

• Rigid grippers for standardized containers

• Flexible grippers and dexterous robotic hands for irregular or fragile objects

• Vacuum suction systems for cartons and sheet materials

2.3.3 Sensor and Energy Systems

Sensor suites typically include:

• 2D/3D cameras

• LiDAR sensors

• Six-axis force sensors

• Torque sensors

• IMUs

Energy systems generally consist of lithium battery packs and Battery Management Systems (BMS), with industrial requirements including:

• Minimum 8-hour operating endurance per shift

• More than 1,500 charge-discharge cycles

The overall development trend is toward higher integration, modularization, and domestic manufacturing capability.

3. Analysis of Core Technology Segments Across the Industry Chain

3.1 Upstream: Core Component Technology Barriers

Core components account for more than 60% of total robot costs and represent the most technically challenging segment of the industry.

AI edge computing chips perform visual processing and foundation model inference, while motion controllers manage closed-loop servo control. Yikong Intelligent has introduced proprietary motion controllers and multi-robot scheduling solutions, enabling effective replacement of imported products in low- and medium-speed industrial applications.

Domestic manufacturers have also achieved significant progress in harmonic reducers, high-precision servo motors, six-axis force sensors, and high-performance 3D LiDAR systems.

3.2 Midstream: Robot Integration and Manufacturing

The midstream segment focuses on robot body design, system integration, assembly, and testing.

Three primary integration approaches are currently adopted:

• Commercial mobile chassis combined with self-developed upper-body systems

• Full-stack in-house development

• Modular component assembly

Key engineering validation tests include vibration testing, load-cycle testing, environmental testing, and long-term reliability verification. Typical MTBF requirements exceed 2,000 operating hours.

3.3 Downstream: Technical Requirements for Manufacturing and Logistics Applications

3.3.1 Warehousing and Logistics

Material Handling

Payload capacities typically range from 10 to 100 kg, with travel speeds up to 1.5 m/s. Heavy-duty chassis control and fleet scheduling are critical technologies.

Intelligent Sorting

Rapid 3D vision recognition and flexible gripping systems enable sorting accuracies exceeding 99.5%, while dynamic force control minimizes product damage.

Inventory Inspection

Autonomous navigation combined with barcode and RFID recognition can achieve identification accuracy above 99.9%, while lifting mechanisms require positioning accuracy within ±2 mm.

3.3.2 Industrial Manufacturing

Machine Loading and Unloading

Repeated positioning accuracy generally must remain within ±0.5 mm.

Equipment Inspection

Multi-sensor fusion technologies support defect detection, instrument reading, and predictive maintenance.

Hazardous Operations

Industrial robots deployed in hazardous areas require dustproof, oil-resistant, and explosion-resistant designs, typically with protection ratings above IP54.

4. Future Development Trends and Engineering Optimization Directions

4.1 Algorithm Development Trends

• Deep lightweight optimization of foundation models for local edge inference

• Digital twin-driven virtual-real hybrid training to reduce deployment costs

• Advanced multi-robot scheduling algorithms for dynamic task allocation and collision avoidance

4.2 Hardware Development Trends

• Deeper mechatronic integration replacing traditional chassis-plus-arm architectures

• Standardized modular interfaces for rapid maintenance and component replacement

• Large-scale production of core components to further reduce system costs

4.3 Motion Control Development Directions

• Deep integration of mobility and manipulation, enabling operation during motion rather than stop-and-work processes

• Wider adoption of adaptive force-control technologies

• Predictive fault diagnosis based on operational data to improve maintenance efficiency and system reliability

5. Conclusion

Wheeled humanoid robots represent the convergence of AGV/AMR mobility technologies and humanoid manipulation systems. Owing to their technological maturity, cost advantages, and operational stability, they are emerging as the most practical embodiment of embodied intelligence for manufacturing and logistics applications.

From a technical perspective, their competitiveness depends on the integrated performance of three interconnected layers: robust perception and decision-making systems driven by foundation models and SLAM technologies; precise motion control systems capable of multi-degree-of-freedom coordination and force control; and reliable actuation systems built upon high-performance core components and advanced integration technologies.

Companies such as Yikong Intelligent are accelerating the transition from prototype development to large-scale deployment through continuous innovation in motion control, multi-robot collaboration, and system calibration technologies.

Over the next three to five years, industry competition is expected to focus on four key areas: model lightweighting, mechatronic integration, fleet collaboration, and cost optimization. For the manufacturing and logistics sectors, wheeled humanoid robots are poised to gradually replace both manual labor and single-function automation equipment, becoming a core enabling technology for flexible intelligent logistics and next-generation automated production systems.