Enquire about Robot

Simply fill up the form, our team will then reach out to you to connect to the relevant parties who can help facilitate your request.

Brand & Model
All fields required. Refer to our privacy policy for questions.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
HomeRobots

AgiBot A2

AgiBot A2 is a 1.69 m bipedal humanoid with 49+ DoF, 350-512 Nm peak torque, GO-1 AI, and tri-market certification for industrial logistics, assembly, and warehousing
Software Type
Closed Source
Software Package
Uses WorkGPT, AgiBot's proprietary multimodal large language model for natural language understanding and task sequencing
Aparobot Readiness Score
ARS?
0
...
Actuators
Quasi-direct drive actuators, High-torque planetary reducers, Peak torque: 350–512 Nm (depending on configuration)
Compiute
200 TOPS AI computing system, real-time control and perception modules operating in parallel.
Sensors
360° Lidar, 6 high-definition cameras, RGB-D camera Fisheye camera, Microphone array Force/torque sensors, Visual-based fingertip sensors
Max Op. time
180
mins

Recent Robot Videos

*Aparobot claims no ownership of videos posted unless otherwise stated.
Description text
Date
  1. https://www.youtube.com/watch?v=NSMw27jlN14, AGIBOT A2 Compeleted Guinness World Records of Longest Journey Walked by a Humanoid Robot, 2025-11-22

Robot Brief

The AgiBot Expedition A2 is a full-size bipedal humanoid robot developed by AgiBot and first deployed in 2024 as the company's flagship general-purpose platform for industrial automation. Designed to execute labour-intensive tasks across logistics, manufacturing, and warehousing environments, the A2 stands 1,690 mm tall and weighs 69 kg with its battery. Built around quasi-direct drive and high-torque planetary actuators delivering peak joint torques of 350 to 512 Nm depending on joint configuration, and equipped with 49 or more active degrees of freedom including 19-DoF dexterous hands, the A2 is certified for operation in China, the United States, and the European Union. It is powered by the GO-1 generalist embodied AI foundation model and the WorkGPT multimodal language model, enabling autonomous manipulation, natural language task instruction, and real-world workflow integration.

Use Cases

  • Autonomous Manipulation and Pick-and-Place:
    Performs precise object grasping, placement, and handling using 19-DoF dexterous hands with visual-based fingertip sensors capable of tasks as fine as threading a needle, across logistics, manufacturing, and retail environments.
  • Industrial Assembly and Quality Inspection:
    Executes structured assembly line tasks, sub-millimetre component handling, and quality inspection duties, replacing manual steps in high-mix, flexible production lines.
  • Autonomous Indoor Navigation:
    Navigates independently through complex factory and warehouse environments using 360-degree LiDAR, six HD cameras, and RGB-D sensors for SLAM-based localisation and real-time obstacle avoidance.
  • Natural Language Task Instruction:
    Interprets spoken task instructions via the WorkGPT multimodal engine, achieving 96% accuracy across text, audio, and visual inputs for context-aware task sequencing and execution.
  • Teleoperation and Embodied AI Data Collection:
    Supports full-body VR teleoperation for human-led demonstration recording, feeding into AgiBot's AIDEA data pipeline for model training and continuous performance improvement.
  • Guinness World Record Distance Walking:
    In November 2025, an A2 unit completed a 106.286 km autonomous walk from Suzhou to Shanghai, certified by Guinness World Records as the longest distance walked by a humanoid robot.

Industries

  • Logistics and Warehousing:
    Automates package sorting, transport, loading, unloading, and inventory management in distribution centres and warehouse facilities.
  • Retail:
    Supports stock replenishment, guided customer assistance, and interactive front-of-house engagement in commercial retail environments.
  • Research and Embodied AI Development:
    Serves as a primary development and benchmarking platform for embodied AI research, providing real-world deployment data for model training and iteration.

Specifications

Length
-
550
mm
Width
-
550
mm
Height (ResT)
-
mm
Height (Stand)
-
1690
mm
Height (Min)
-
mm
Height (Max)
-
1690
mm
Weight (With Batt.)
-
69
kg
Weight (NO Batt.)
-
55
kg
Max Step Height
-
mm
Max Slope
+/-
-
°
Op. Temp (min)
-
°C
Op. Temp (Max)
-
°C
Ingress Rating
-
No items found.

Intro

The AgiBot Expedition A2 stands 1,690 mm tall with a footprint of 550 mm (L) x 550 mm (W) and weighs 69 kg with its 700 Wh battery installed, or 55 kg without. It supports a battery swap system and delivers approximately 2 hours of runtime per charge. The robot's actuation system consists of quasi-direct drive actuators for high-bandwidth, compliant joint control and high-torque planetary reducers for load-bearing joints, with peak joint torques ranging from 350 Nm to 512 Nm depending on the joint configuration. This torque range enables the A2 to handle payloads beyond what most comparable humanoids support, with a heavy-duty Expedition A2-Max variant capable of carrying up to 40 kg. The A2 Ultra configuration provides 49 or more active degrees of freedom, including 7 DoF per arm and 19 DoF per dexterous hand, enabling manipulation of objects as precise as threading a needle. A six-axis force/torque sensor is integrated into each arm, alongside visual-based fingertip sensors that combine tactile and visual feedback for fine force-regulated grasping.

The perception system comprises a 360-degree LiDAR, six high-definition cameras providing wide-field and stereo coverage, an RGB-D depth camera, fisheye cameras for peripheral coverage, and a microphone array for voice interaction. The A2's compute architecture provides 200 TOPS of AI processing for real-time perception, planning, and control. It is powered by WorkGPT, AgiBot's proprietary multimodal language model, which processes text, audio, and visual inputs with a reported 96% accuracy, enabling natural language task instruction and context-aware multi-step task execution. The GO-1 generalist foundation model governs manipulation behaviour. A VR Teleoperation Kit is available as an optional accessory, enabling full-body remote operation for human demonstration data collection. The A2 received safety certification for China, the United States, and the European Union in May 2025, making it one of the few humanoid robots to achieve tri-region regulatory compliance at the time of this audit.

Connectivity

  • WiFi connectivity (2.4 GHz and 5 GHz bands)
  • Ethernet port for wired network access
  • USB and HDMI ports for peripherals
  • Bluetooth for short-range device communication
  • Cloud connectivity for remote monitoring, diagnostics, and software updates.

Capabilities

  • 19-DoF Dexterous Hands with Fingertip Sensing:
    Each hand provides 19 degrees of freedom with integrated visual-based fingertip sensors, enabling grasping of small, fragile, and geometrically complex objects with force-regulated precision.
  • WorkGPT Multimodal Task Instruction:
    AgiBot's proprietary multimodal model processes voice, vision, and text inputs to generate and execute multi-step task plans from natural language instructions, achieving reported 96% cross-modal accuracy.
  • GO-1 Generalist Foundation Model:
    Governs real-world manipulation behaviour, enabling zero-shot and few-shot task generalisation across object types and environments not specifically pre-programmed.
  • 360-Degree Perception and Autonomous Navigation:
    Fused LiDAR, RGB-D, fisheye, and HD camera data enables full-coverage environmental awareness and SLAM-based autonomous navigation in dynamic factory and warehouse environments.
  • Battery Swap and Sustained Operation:
    Supports hot-swappable battery replacement to minimise downtime, with a 700 Wh capacity providing approximately 2 hours of runtime per pack.
  • VR Teleoperation and Embodied Data Collection:
    Optional VR Teleoperation Kit enables millisecond-latency full-body human-led operation for demonstration data capture, feeding directly into AgiBot's AIDEA training pipeline.