AgiBot A3 - Robot Details, Use Case and Specifications

AgiBot A3

AgiBot Expedition A3 is a 55 kg bipedal humanoid with 10-hour battery, UWB swarm sync for 100-robot performances, and high-agility martial arts capability for entertainment and retail deployment.

Enquire about robot



Data verified by manufacturer

Made by

AgiBot Innovation (Shanghai) Technology Co., Ltd.

Software Type

Closed Source

Software Package

GCFM (Generative Control Foundation Model) for convert text, audio, or video inputs into real-time natural robot motion sequences. BFM (Behavioural Foundation Model), enables single-demonstration movement imitation for rapid choreography deployment. WITA Omni, AgiBot's end-to-end multimodal interaction model fusing vision, audio, language, and action. WorkGPT: multimodal large language model for natural language understanding and task sequencing. Link-U OS: AgiBot's native operating system for embodied intelligence. LinkSoul Platform: persistent personality, memory, and long-term interaction engine

...

Report

Actuators

Compiute

Sensors

Equipped with 360-degree multi-array microphone system, shoulder tactile sensors.

Max Op. time

600

mins

Recent Robot Videos

*Aparobot claims no ownership of videos posted unless otherwise stated.

Description text

Date

https://www.youtube.com/watch?v=P8_GKWXWGP4, AgiBot A3 Showreel, 2026-05-10

Robot Brief

The AgiBot A3 is a full-size bipedal humanoid robot developed by AgiBot and unveiled in February 2026 as the company's high-performance interactive service and entertainment platform. Positioned as a "silicon-based stage star," the A3 stands 1,730 mm tall and weighs 55 kg, constructed from lightweight magnesium, titanium, and TPU materials to achieve an industry-leading power-to-weight ratio of 0.218 kW/kg. Equipped with 10 hours of continuous battery runtime, a 10-second hot-swap battery exchange system, UWB centimetre-level swarm positioning enabling synchronised multi-robot performances of up to 100 units, shoulder tactile sensing, and a 360-degree multi-array microphone system, the A3 is designed for high-frequency audience-facing environments including entertainment performances, brand activations, retail showrooms, and large-scale exhibitions. Its athletic capability, including verified aerial kicks, mid-air manoeuvres, backflips, and rapid spinning movements, is produced by high-power-density actuators, advanced balance algorithms, and an end-to-end AI stack running on the GCFM generative motion model and WITA Omni multimodal interaction model.

Use Cases

High-Agility Dynamic Motion Performances:
Executes martial-arts-style aerial manoeuvres including flying kicks, consecutive mid-air strikes, cyclone kicks, and backflips in real-world conditions, verified by AgiBot as filmed without CGI.
Synchronised Multi-Robot Choreography:
Uses UWB centimetre-level swarm positioning to coordinate simultaneous performances of up to 100 A3 units in precisely timed, spatially synchronised group routines for live events and brand activations.
Natural Language and Multimodal Customer Interaction:
Engages audiences and customers in natural conversation using the WITA Omni multimodal interaction model, which fuses vision, audio, language, and motion for contextually aware, emotionally expressive responses.
Retail Guidance and Showroom Assistance:
Functions as an interactive product guide and customer engagement agent in retail and commercial environments, combining physical presence, natural speech, and human-recognisable gestures.
Live Entertainment and Stage Performance:
Performs choreographed dance routines, theatrical movements, and sustained high-intensity physical sequences on stage, validated through the AgiBot Night 60-minute live gala event in February 2026.
Wake-Word-Free Conversational Interaction:
An end-to-end large AI model enables natural conversation initiation from voice without requiring wake-word activation, and responds to shoulder-tap gestures, supporting organic interaction flows in public settings.

Industries

Entertainment and Live Events:
Designed as the primary performance robot for concerts, galas, brand launch events, and large-scale live shows requiring synchronised multi-robot choreography and dynamic physical performance.
Retail and Commercial Spaces:
Deployed as an interactive guide and brand ambassador in showrooms, flagship stores, and commercial centres, combining physical agility with natural language customer engagement.
Hospitality and Tourism:
Serves as an experiential interaction point in hotels, theme parks, museums, and tourist attractions where memorable, dynamic human-robot interaction is a primary draw.
Corporate Events and Exhibitions:
Functions as a demonstration centrepiece at trade shows, product launches, and exhibitions, attracting and engaging audiences through live performance and interactive dialogue.
Education:
Applicable in educational settings for STEM demonstration, physical AI showcase, and student interaction with an expressive, approachable humanoid platform.

Specifications

Length

Width

Height (ResT)

Height (Stand)

1730

Height (Min)

Height (Max)

Weight (With Batt.)

Weight (NO Batt.)

Max Step Height

Max Slope

+/-

Op. Temp (min)

°C

Op. Temp (Max)

°C

Ingress Rating

No items found.

Intro

The AgiBot Expedition A3 stands 1,730 mm tall and weighs 55 kg, built using a lightweight combination of magnesium, titanium, and TPU structural materials that yield a power-to-weight ratio of 0.218 kW/kg, which AgiBot states is the highest in its product class at the time of launch. Its leg structure adopts a lightweight exoskeleton-inspired design that improves dynamic stability and agility during high-speed and airborne motion sequences. The waist joint provides a full range of rotational and lateral motion that closely mirrors the human torso, enabling the fluid trunk movements required for martial arts-style manoeuvres and dance performances. The full-body degrees of freedom breakdown per limb segment has not been publicly disclosed, but the body is described by AgiBot as incorporating highly anthropomorphic full-body articulation across all segments. All joints are driven by high-power-density actuators capable of explosive torque output without destabilising the platform, and the control system integrates high-frequency balance algorithms for real-time centre-of-mass adjustment during transient airborne phases.

The A3 operates for up to 10 hours on a single battery pack, the longest runtime in AgiBot's bipedal humanoid lineup, with a hot-swap battery exchange system completing a full replacement in 10 seconds. For fleet operation, the A3 integrates UWB (Ultra-Wideband) radio positioning to centimetre-level accuracy, enabling up to 100 A3 units to execute precisely coordinated choreography in a shared physical space. Shoulder tactile sensors detect physical contact and enable reactive responses, while a 360-degree multi-array microphone system captures voice input from any direction without requiring the interacting person to face the robot directly. The AI stack runs the GCFM (Generative Control Foundation Model), which converts text, audio, or video inputs into real-time physical motion sequences, and the WITA Omni model, which delivers context-aware, emotionally expressive multimodal interaction by fusing vision, audio, language, and physical action. The BFM (Behavioural Foundation Model) enables the A3 to imitate new human movements from a single demonstration video, allowing rapid deployment of new choreographic routines without prolonged training cycles.

Connectivity

WiFi Supported
Standard UWB (Ultra-Wideband), for centimetre-level swarm positioning across up to 100 units
OTA Supported for model and firmware updates

Capabilities

High-Power-Density Actuators for Dynamic Motion:
Joints are driven by actuators optimised for explosive torque delivery, enabling the simultaneous multi-joint coordination required for aerial kicks, rapid spinning, and mid-air stabilisation manoeuvres.
UWB Swarm Positioning at Centimetre Accuracy:
Integrated UWB radio enables sub-centimetre spatial awareness across fleets of up to 100 A3 units, enabling large-scale synchronised group performances in real physical environments.
GCFM Generative Motion Control:
The Generative Control Foundation Model converts text, audio, or video prompts into real-time natural movement sequences, enabling dynamic improvisation and rapid deployment of new performance routines.
WITA Omni Multimodal Interaction:
An end-to-end robot-native interaction model fuses vision, audio, language, and physical action to deliver emotionally intelligent, context-aware responses with synchronised speech, gestures, and expressions.
BFM Single-Shot Movement Imitation:
The Behavioural Foundation Model enables the A3 to learn and reproduce a new human movement from a single demonstration or short video, dramatically reducing deployment time for new choreographic requirements.
10-Second Hot-Swap Battery Exchange:
The battery replacement system completes a full pack swap in 10 seconds, enabling continuous performance scheduling across multi-hour event programmes without production downtime.