Glossary
Terms used in robot learning hardware and data collection, in plain words.
- ACT
- Action Chunking with Transformers: an imitation-learning method that predicts short sequences of future actions from camera images and joint states. ARX's open-source data pipeline is migrated from ACT.
- ALOHA
- A low-cost bimanual teleoperation setup from Stanford with two leader and two follower arms. Mobile ALOHA put it on a wheeled base. ARX's data-collection repositories are adapted from mobile-aloha and act-plus-plus.
- Bimanual manipulation
- Tasks done with two arms at once, such as folding cloth or handing an object from one hand to the other.
- CAN bus
- Controller Area Network: the wired bus ARX arms use to talk to the computer, through a USB-CAN or EtherCAT-CAN adapter.
- Degrees of freedom (DoF)
- The number of independent joints an arm can move. A 6-DoF arm can place its gripper at any position and orientation in its workspace.
- Egocentric data
- Video and motion recorded from the operator's own point of view, often with a head-mounted or handheld camera.
- End effector
- The tool at the end of an arm, usually a gripper.
- Force control
- Controlling an arm by the force or torque at each joint, not only by position. It lets an arm comply on contact and be guided by hand.
- Gravity compensation
- A control mode that cancels the arm's own weight so it can be moved by hand and stays where it is left.
- Imitation learning
- Training a robot policy from demonstrations of a task, usually recorded by teleoperation.
- Leader–follower teleoperation
- A person moves a leader arm by hand and a follower arm copies the motion joint by joint. The recorded motion and camera video become training data.
- MIT mode
- A motor command that sets stiffness (kp), damping (kd), target position, target velocity and feed-forward torque for a joint at once. The ARX A5 SDK exposes it.
- Mobile manipulation
- Manipulation done by arms on a moving base, so the robot can work across a room.
- Payload
- The mass an arm can carry at its end effector. ARX X5: 2 kg rated, 3 kg peak.
- Policy
- The trained model that maps observations, such as images and joint states, to robot actions.
- Quasi-direct drive (QDD)
- Motors with a low gear ratio. They keep the arm light and back-drivable and make joint torque easier to sense.
- ROS 2
- Robot Operating System 2, a set of open-source robotics libraries. ARX's collect–train–deploy pipeline runs on ROS 2 (Jazzy on Ubuntu 24.04, Humble on 22.04).
- Teach and replay
- Recording a motion by guiding the arm by hand, then playing it back. arx5-sdk includes a teach-and-replay example for X5.
- Teleoperation
- Controlling a robot remotely in real time: with leader arms, a VR headset, a SpaceMouse or a keyboard.
- UMI
- Universal Manipulation Interface: handheld grippers for recording demonstrations without a robot, from Stanford and Columbia. Its deployment code supports ARX arms.
- VLA (vision-language-action model)
- A model that takes camera images and a language instruction and outputs robot actions. π0 is an example.
- Working radius
- How far from its base an arm can reach. ARX X5: 620 mm.
- World model
- A model that predicts how a scene will change in response to actions, used for planning or for training policies.
Get a price for your setup.
Price, lead time and shipping options for your country.