DOBOT Atom D Data Collection Humanoid Robot

• DOBOT Atom D Data Collection Humanoid Robot
• Compact upper-body humanoid platform for large-scale embodied AI data collection
• 16 total DoF with dual 7-DoF arms and a 2-DoF head for manipulation-focused data workflows
• ±0.05mm repeat positioning accuracy and 3.5kg rated payload per arm
• Full HD binocular camera and Intel RealSense D455 depth sensor for high-quality visual data capture
• Intel i5 basic computing module; Ethernet connectivity for external data pipelines
• Includes 2-hour swappable battery pack; optional mobile base, wrist cameras, and end effectors

The DOBOT Atom D Data Collection Humanoid Robot is a compact upper-body humanoid robotic platform developed by DOBOT and purpose-built for large-scale embodied AI data collection. Rather than a full-body bipedal robot, the Atom-D focuses exclusively on the arms and head, delivering the same dual seven-degree-of-freedom arm precision found in the full Atom series within a smaller, stationary form factor suited to desktop-level and lab environments. Its binocular camera system is tuned to minimize VR and MR motion artifacts during teleoperation sessions, ensuring high-quality visual recordings for imitation learning pipelines. The platform is designed for research teams and institutions that prioritize the volume and quality of training data over full-body locomotion capabilities.

The standard Atom-D connects via Ethernet for streamlined data transfer to external collection and processing infrastructure, making it well-suited for high-throughput demonstration recording workflows. Optional wrist-mounted RGB-D cameras, dexterous hands, and a teleoperation kit are available separately for teams that want to expand data collection capabilities over time. An integrated swappable battery pack supports untethered sessions, and an optional mobile base extends the working area without requiring full bipedal locomotion hardware. With full SDK access and a modular optional ecosystem, the Atom-D provides a cost-effective entry point into production-grade manipulation data collection for embodied AI research programs.

Why Buy From RobotShop

Authorized Distributor

Full manufacturer-backed warranty coverage and dedicated product support

Hassle-Free Delivery

Complete end-to-end logistics management, including customs clearance

Single Point of Contact

All pre-sales and post-sales support handled by our dedicated team

Documentation & Resources

Full access to detailed technical guides and a comprehensive knowledge base

Data Collection Humanoid Robot

DOBOT Atom D

Compact dual-arm data collection platform with binocular vision and Ethernet connectivity for high-throughput AI training datasets.

Platform Features

The DOBOT Atom-D delivers industrial-grade dual-arm precision and high-fidelity binocular vision in a compact, stationary upper-body form factor purpose-built for AI data collection workflows.

Bionic
Arms →

Bionic
Arms

7-DoF per Arm

Single arm DoF: 7

Payload: 3.5 kg

Reach: 600 mm

Repeat accuracy: ±0.05 mm

Max end speed: 1.5 m/s

Vision
System →

Vision
System

Dual Head Cameras

Binocular camera: Full HD, 60fps

Depth camera: Intel RealSense D455

Depth range: Up to 6m

Motion artifact: Minimized for VR/MR teleoperation

Optional wrist-mounted RGB-D cameras available for close-range manipulation capture.

Data
Architecture →

Data
Architecture

Chip Label

Connectivity: Ethernet

Computing: Intel i5 base module

Power: 2h swappable battery + AC charger

SDK: Supported

Designed for integration with external data collection and processing infrastructure.

Capabilities

The DOBOT Atom-D combines industrial-precision dual arms with a high-fidelity binocular vision system and flexible data infrastructure to support scalable embodied AI data collection programs.

Precision Dual-Arm Manipulation

Dual seven-degree-of-freedom arms share the same architecture as the full Atom series, delivering industrial-grade repeatability for consistent, high-quality demonstration recording. Both arms can handle tools, objects, and fixtures across a range of manipulation task types for imitation learning.

High-Fidelity Binocular Vision

A 60FPS Full HD binocular camera tuned to minimize VR and MR motion artifacts captures consistent visual data during teleoperation sessions. An Intel RealSense D455 with a 6-meter depth range provides structured RGB-D data for training perception-aware manipulation models.

Ethernet Data Pipeline

Ethernet connectivity links the Atom-D directly to external collection and processing infrastructure, enabling efficient transfer of large demonstration datasets without relying on onboard storage. This makes it straightforward to integrate into existing data management and annotation pipelines.

Modular Expansion Options

Optional wrist-mounted RGB-D cameras capture close-range end-effector data for detailed manipulation studies. An optional mobile base extends the working area without requiring bipedal locomotion hardware, and optional dexterous hands or 2-finger grippers can be added as programs scale.

Use Cases & Application Scenarios

The DOBOT Atom-D EDU is built for teams that need a fully-bundled, lab-friendly platform for recording large volumes of high-quality manipulation demonstrations, without the footprint or complexity of a full bipedal humanoid and without sourcing accessories separately.

Turnkey Manipulation Data Collection

Research teams use the Atom-D EDU to record high-volume demonstration datasets for imitation learning and behavior cloning, with the dexterous hand, wrist cameras, teleoperation kit, and AI compute module all included from day one. The configuration removes the typical weeks of accessory procurement and integration work that slow down data collection programs.

VR Teleoperation & Demonstration Recording

The bundled VR teleoperation kit and motion-artifact-tuned binocular camera let operators run fluid teleoperation sessions with less fatigue, producing cleaner and more consistent demonstrations. Sessions can be recorded across a range of task types and object configurations to build diverse training sets.

Classroom & Lab AI Training Programs

Universities and embodied AI courses can deploy the Atom-D EDU as a compact, shared training platform that fits standard lab benches. Because it ships ready to use, instructors can focus class time on embodied AI concepts instead of hardware integration and accessory setup.

Edge-Inference Policy Development

The onboard 1500 TOPS AI module lets teams train and validate manipulation policies with real-time inference directly on the platform, without routing through external GPU servers. This shortens iteration cycles when developing task-specific models that will later transfer to full humanoid platforms sharing the same arm architecture.

Multi-Unit Data Infrastructure

Organizations building large-scale embodied AI datasets can run multiple Atom-D EDU units in parallel, connected over Ethernet to centralized collection infrastructure. The bundled, standardized configuration simplifies deployment across labs where consistent hardware setup across stations is required.