AstroD. AD-01
Foundation

AstroD. AD-01

AstroDroid AD-01 brings embodied AI from concept to floor operations. Powered by Hyper‑VLA for ~200 ms end‑to‑end response, 11‑DOF dexterous hands, and a wheeled base that excels indoors. Modular, ROS‑compatible and cloud‑updatable—built for manufacturing, logistics and retail. Advanced prototype; pilot programs in 2025.

Description

AstroDroid AD-01, developed by INFIFORCE under the ELU.AI Technology Group, marks a significant advancement in embodied AI humanoid robotics. Unveiled on September 24, 2025, at the Alibaba APSARA Conference in Hangzhou, this advanced prototype represents the company's 'One Brain, Multiple Bodies, Multiple Scenarios' vision. Standing at 170 cm tall and weighing 120 kg, AD-01 features a hybrid wheeled base for efficient indoor navigation (max speed 7.2 km/h) combined with a humanoid upper body equipped with 11-DOF dexterous hands (22 DOF total for arms/hands, contributing to overall 43 degrees of freedom). The hands support precision grasping up to 3 kg payload with force-torque feedback for human-like adaptability in tasks ranging from delicate assembly to material handling. At its core is the proprietary Hyper-VLA (Vision-Language-Action) end-to-end multimodal model, enabling ~200 ms glass-to-action latency. This full-stack self-developed AI integrates natural language understanding, real-time scene perception via trinocular vision, dual LiDAR, and a multimodal sensor matrix, and autonomous action generation. The cloud-edge collaborative architecture allows continuous learning, remote updates, and expansion of capabilities. Hardware includes NVIDIA Jetson AGX Orin 64GB for edge computing, high-precision BLDC motors with harmonic drive and planetary gears for low-backlash transmission, and an aluminum alloy frame with carbon fiber composites. IP42 ingress protection ensures suitability for indoor industrial environments. ROS2 compatibility and Linux/Ubuntu OS facilitate integration. Designed primarily for manufacturing, logistics, and retail, AD-01 excels in floor operations like picking, sorting, and inspection. Pilot programs commenced in late 2025, with no widespread commercial deployments yet reported. The development team, boasting 70%+ PhDs/Masters from Tsinghua, Zhejiang University, Stanford, and alumni from Alibaba, Huawei, Baidu, has secured awards like BEYOND Expo Best Choice and Hangzhou Major Innovation Projects. ELU.AI, founded by ex-Alibaba SVP Isabella Bai, focuses on embodied intelligence, with INFIFORCE handling robotics. While still in prototype stage (priced ~$80,000-$100,000), AD-01 demonstrates demos in household chores, caregiving, cooking, cleaning—hinting at broader applications. Runtime is 2 hours per charge, with modular design for future bipedal upgrades. This positions AD-01 as a bridge from research to practical AI robotics, competing in China's booming humanoid market backed by government funds.

Key Features

Hyper-VLA Multimodal AI

Proprietary end-to-end Vision-Language-Action model with <200ms latency, enabling natural language instructions, real-time perception, and autonomous action execution.

Dexterous 11-DOF Hands

Precision manipulation with force-torque sensing, 10 fingers, supporting 3kg payload for assembly, handling, and versatile grasping.

Hybrid Wheeled Mobility

Indoor-optimized base reaches 7.2 km/h, combining efficient navigation with upper-body dexterity for industrial floors.

Advanced Sensing Suite

Trinocular vision, dual LiDAR, multimodal sensors for robust environmental perception and safe human interaction.

Modular ROS-Compatible Design

Cloud-edge updates, NVIDIA Jetson AGX Orin, ROS2 integration for seamless deployment and scalability.

Specifications

AvailabilityPrototype
NationalityChina
Websitehttp://www.elu-ai.com
Degrees Of Freedom, Hands11
Height [Cm]170
Manipulation Performance3
Navigation Performance2
Max Speed (Km/H)7.2
Strength [Kg]3
Weight [Kg]120
Runtime Pr Charge (Hours)2
Safe With HumansYes
Cpu/GpuNVIDIA Jetson AGX Orin 64GB
Ingress ProtectionIP42
Connectivity– Wi-Fi 6 (802.11ax) – Ethernet (Gigabit) – 4G/5G Optional – Bluetooth 5.0 – ROS Compatible
Operating SystemLinux-based (Ubuntu 20.04/22.04) ROS 2 (Robot Operating System) Custom AI Runtime Environment
Llm IntegrationCloud-edge collaborative architecture, Integrated with Hyper-VLA Multimodal Model, Natural language understanding and instruction following, Proprietary vision-language-action model, Real-time multimodal reasoning, Yes
Latency Glass To Action<200ms (End-to-End)
Motor TechHigh-precision Brushless DC Motors (BLDC) Integrated Joint Actuators with Position/Torque Control
Gear TechHarmonic Drive Reducers Planetary Gear Systems High-precision Transmission with Low Backlash
Main Structural MaterialAluminum Alloy Frame Carbon Fiber Composite Components Engineering Plastics for Housing
Number Of Fingers10
Main MarketIndustries, logistics, Manufacturing, Research & education
VerifiedNot verified
Walking Speed [Km/H]2
ManufacturerFoundation
Height Cm170
Weight Kg120
Dof Total43
Dof Hands11 per hand
Payload Kg3
Max Speed Kmh7.2
Runtime Hours2
Cpu GpuNVIDIA Jetson AGX Orin 64GB
OsLinux (Ubuntu 20.04/22.04), ROS 2
SensorsTrinocular vision, dual LiDAR, multimodal sensor matrix, force-torque feedback
ConnectivityWi-Fi 6 (802.11ax), Gigabit Ethernet, optional 4G/5G, Bluetooth 5.0
MotorsHigh-precision BLDC with position/torque control
GearsHarmonic Drive reducers, planetary gear systems (low backlash)
MaterialsAluminum alloy frame, carbon fiber composites, engineering plastics
Ip RatingIP42
Ai ModelHyper-VLA (cloud-edge collaborative)
Latency Ms<200 (end-to-end)

Curated Videos

Frequently Asked Questions

What is the primary AI architecture in AstroD. AD-01?

It uses ELU.AI's self-developed Hyper-VLA end-to-end multimodal model, integrating vision, language, and action for real-time reasoning, natural language processing, and <200ms latency in complex environments like factories or retail spaces.

What are the mobility and manipulation capabilities?

Wheeled base enables 7.2 km/h max speed indoors; 11-DOF hands with 3kg strength handle precise tasks. Total 43 DOF, with force feedback for adaptive grasping across diverse objects.

Is it safe for human collaboration?

Yes, IP42 protected, multimodal sensors including LiDAR ensure collision avoidance and safe operation around humans in shared workspaces like manufacturing or logistics.

What is the battery life and computing hardware?

Provides 2 hours runtime per charge on NVIDIA Jetson AGX Orin 64GB edge compute, supplemented by cloud collaboration for heavy AI processing and OTA updates.

What markets is it targeted for and availability?

Focused on manufacturing, logistics, retail; advanced prototype stage with 2025 pilot programs. Modular design supports research, custom development, and future expansions.

How does it integrate with existing systems?

ROS2 compatible on Ubuntu Linux, Wi-Fi 6, Gigabit Ethernet, optional 5G; open architecture allows easy integration into automation pipelines and fleet management.

×