7-AXIS / RESEARCH MANIPULATOR
A lightweight 7-DOF arm for fixed-workstation manipulation and robot learning. Redundant kinematics, hand-guided teaching and hybrid force-position control support research on obstacle-aware planning, demonstration capture, impedance control and imitation learning. Choose a left arm, right arm or dual-arm set per task.
Data outputs, interface adaptation and delivery scope are confirmed per task.
The anthropomorphic 7-axis configuration opens up research on end-effector pose holding, joint obstacle avoidance and redundancy resolution. At 4.3 kg per arm with a rated repeatability of ±0.1 mm, it suits lab benches and fixed workstations.
The manual lists gravity compensation, zero-force hand guiding and trajectory playback, with a 100 Hz teaching sampling rate, recording joint position, velocity and torque-related data. Sampling rate is not control frequency, and torque readings are not end-effector contact force.
Joints support position, velocity, torque and MIT hybrid control. Developer resources cover the Python / C++ SDK, ROS 2 Humble, MoveIt2, Gazebo and URDF; driver versions, interfaces and examples are confirmed per project.
ATLAS A7 / IN THREE DIMENSIONS
Drag to inspect the shoulder, elbow, wrist and end interface, switch between front, side and back, and zoom in on A7 structural details.
DIMENSIONS & INTERFACES
Original brochure dimension drawings, viewed separately for the whole arm, shoulder, end and joint numbering. The product exterior is gray-blue; engineering drawings keep their original annotations.
Source: 7-axis arm product manual, dimension drawing page (PAGE 04). This page supports selection and integration discussions; use the formal engineering drawings and models for your version for machining, mounting and collision checks, and do not measure from web images.
READY FOR YOUR WORKBENCH
Confirm mounting and power first, then drivers, zero points and teaching records, building reproducible manipulation experiments step by step.
Select per task and check the shoulder mounting surface, end interface and workspace. Rated weight is 4.3 kg per arm and rated max end payload is 3 kg.
The manual requires a supply of at least 6 A. CAN runs at 1 Mbps and connects to the workstation via USB-to-CAN; adapter and cables are supplied per the configuration list.
Ubuntu 22.04 LTS recommended. The toolchain includes the Python / C++ SDK, MoveIt2, Gazebo and URDF, with versions and license scope confirmed per project.
Hand-guide the arm at low speed with zero-force control and gravity compensation, recording joint position, velocity and torque-related data, with trajectory playback. Sampling rate is not control frequency.
Make sure the range of motion is clear and check zero points and safety boundaries. Hand-guided teaching requires zero-force control to be enabled correctly; never force the arm when powered off or with brakes engaged. High-speed or contact tasks need end-effector protection and risk controls validated first.
THE DETAILS MATTER
Key Atlas A7 specifications below; choose left, right or dual-arm configurations.
| Parameter | Spec / configuration |
|---|---|
| Degrees of freedom | 7 axes (7-DOF) |
| Weight per arm | 4.3 kg |
| Max end-effector payload | 3 kg |
| Max reach | 739.45 mm |
| Repeatability | ±0.1 mm |
| Materials | Aluminum alloy frame, plastic housing |
| Mounting | Fixed workstation; mounting surface must be level and rigid, hole pattern per the drawing for your version |
| Options | Left arm / right arm / dual-arm set |
Max payload is not a guarantee at every pose; end-tool mass, center of gravity and dynamic conditions must be included in selection. Repeatability is not absolute accuracy.
| Joint | Range of motion | Max speed | Rated torque | Peak torque |
|---|---|---|---|---|
| J1 | -60° to +180° | 180°/s | 15 N·m | 30 N·m |
| J2 | -180° to +15° | 180°/s | 15 N·m | 30 N·m |
| J3 | ±158° | 180°/s | 15 N·m | 30 N·m |
| J4 | ±110° | 180°/s | 15 N·m | 30 N·m |
| J5 | ±158° | 180°/s | 3 N·m | 7 N·m |
| J6 | ±90° | 180°/s | 3 N·m | 7 N·m |
| J7 | ±158° | 180°/s | 3 N·m | 7 N·m |
J1–J4 are the upper-arm joints and J5–J7 the wrist joints. Values are from the manual; coordinates, zero points and sign conventions for left and right arms follow the corresponding driver and mounting version. Peak torque is not continuous output.
| Parameter | Spec / configuration |
|---|---|
| Power supply | DC 48 V; the manual requires a supply of at least 6 A, with headroom for actual operating conditions |
| Total power | <300 W |
| Communication | CAN 2.0, 1 Mbps; via the supplied USB-to-CAN adapter |
| External ports | Power port × 1, CAN port × 1 |
| Joint encoders | 21-bit dual encoders |
| Joint control modes | Position / velocity / torque / MIT hybrid control |
| Teaching | Gravity compensation, zero-force hand-guided recording, trajectory playback |
| Teaching sampling rate | 100 Hz |
| Teaching records | Joint position, velocity and torque-related data; fields, units and timestamps checked against the SDK version |
The joint modules' wide voltage range does not replace the system's DC 48 V supply requirement. Neither the CAN bus rate nor the teaching sampling rate equals policy inference or end-to-end control frequency.
| Parameter | Spec / configuration |
|---|---|
| Recommended development environment | Ubuntu 22.04 LTS |
| Development interfaces | Python / C++ SDK; specific APIs and examples per the delivered version |
| Robot middleware | ROS 2 Humble driver package |
| Motion planning | MoveIt2 |
| Simulation and models | Gazebo, URDF and controller configs; models must match the physical robot version |
| Engineering documentation | Mounting and end-effector interface documentation available per project, with version and license confirmed |
| Data Engine integration | Vision, joint state, action commands and task events can be paired per project, with sync and export validated separately |
| Training and inference | Can be paired with an external workstation or cloud service; no built-in GPU, model or pre-trained skills by default |
Available versions and licenses of software, models and drawings are checked item by item. Third-party developer resources are not presented as our own work, and a complete open-source project is not delivered by default.
| Parameter | Spec / configuration |
|---|---|
| Operating temperature | 5–50 °C |
| Relative humidity | 35%–80% RH, non-condensing |
| Noise | <60 dB (measurement conditions not stated in the manual) |
| Arm options | Single left arm, single right arm or dual-arm set; quantity and mounting direction specified in the order |
| Accessories | USB-to-CAN module, USB cable, power and CAN cables, mounting fasteners and tools, delivered per the configuration list |
| Confirmed separately | Whether a power adapter, gripper, tactile end effector, cameras, fixtures, workstation and safety components are included |
FROM CAPTURE TO ARCHIVE
Decide single or dual arm, end-effector mass and center of gravity, workstation space and safety boundaries; check mounting rigidity and power.
Check zero points, drivers and gravity compensation, and capture demonstrations under safe conditions; when vision is used, record calibration and timing alongside.
Validate slow-speed trajectory playback first, then evaluate policies on agreed tasks, keeping human takeover, failure and exception logs.
BUILD YOUR NEXT SETUP
MODULAR / ROBOT LEARNING
Modular platform
eHand-6DEXTEROUS HAND / 6 DOF
Configured per project
Tactile Data GloveTACTILE / ELECTROMAGNETIC TRACKING
Multiple configurationsSTART WITH YOUR TASK
From device selection to data and training workflows, we'll help you find the right place to start.