Articulated object

• Category: Planning & manipulation › Manipulation
• Tags: manipulation, 3d-perception, kinematics, mobile-manipulation, household-robotics

An object made of rigid parts connected by joints, usually revolute or prismatic, such as a door, drawer, laptop or oven, whose parts move relative to each other when a robot manipulates it.

Definition

An articulated object consists of rigid links connected by joints that allow constrained relative motion. In robot manipulation the joints are usually single degree of freedom: revolute joints for doors, lids and laptop screens, and prismatic joints for drawers and sliding panels. Many formulations, including FlowBot3D's, assume that one link is fixed to the world, as with a cabinet body or an oven, so the task is to move the other parts. In general, articulated objects such as scissors, pliers or a laptop on a desk can also move as a whole.

Manipulating such objects means respecting the kinematic constraint. A robot opening a door must move its hand along an arc around the hinge axis; pulling straight back generates large internal forces or loses the grasp. Approaches differ in how much they model. Model-based pipelines estimate the kinematic structure explicitly: which parts are links, the joint type, the axis direction and position, and the current joint state, often by watching parts move and fitting revolute or prismatic models. Learning-based approaches predict this structure from a single point cloud or image, or skip it and predict local quantities directly, such as an articulation flow giving the direction each point on a part would move, or an affordance map of where to grasp and pull. Interactive perception combines the two: the robot moves a part a little, observes the result and updates its estimate.

Articulated objects are usually represented in the same way as robots, as kinematic trees in formats such as URDF. Simulation datasets such as PartNet-Mobility, used with the SAPIEN simulator, provide a large set of annotated household objects for training and testing. Practical difficulties include unknown joint limits, friction and spring loads, latches and handles that need a separate action, and partial views in which the joint axis is hidden.

Why it matters

Much of the everyday built environment is articulated: doors, drawers, cabinets, refrigerators and appliances. Home and service robots must open and close these reliably, and that requires reasoning about kinematic constraints instead of treating the target as a free rigid body.

Key applications

• Opening doors, drawers and cabinets in homes and offices
• Operating appliances such as ovens, dishwashers and refrigerators
• Household mobile manipulation benchmarks
• Building interactive digital twins of indoor scenes
• Using hinged tools such as pliers and scissors

In practice

A team programming a mobile manipulator to open kitchen cabinets captures a point cloud of the cabinet with a depth camera. A network segments the door and predicts either the hinge axis or a per-point motion direction, and the system chooses a grasp on the handle or the door edge. The arm then follows the predicted arc under impedance control, so that errors in the axis estimate are absorbed by compliance instead of building up large forces. Once the door has moved slightly, the system re-estimates the axis from the hand's actual path and continues with the corrected arc, stopping when the measured force indicates the joint limit.

Common misconceptions

• Opening a door is a pure position-control problem — the hand must follow the joint's constraint, so compliance or force control matters as much as accurate positions
• The joint axis must be known exactly before acting — many systems start from a rough estimate and refine it during the interaction
• Articulated objects are always fixed to the world — many benchmarks assume a fixed base for simplicity, but tools and portable devices move as a whole too

Who should know this

• Mobile and household manipulation engineers
• Robot perception researchers working on 3D scene understanding
• Simulation and dataset builders
• Service robot developers

Related technologies

• PartNet-Mobility Dataset
• SAPIEN Simulator
• URDF
• Point Cloud Segmentation Networks
• Kinematic Structure Estimation

Related terms

• Interactive Perception
• URDF
• Impedance Control
• Point Cloud
• Joint
• Affordance

Related projects

• LeRobot: HuggingFace ML library for real-world robotics in PyTorch
• Isaac Lab: NVIDIA robot learning framework: one task API across Newton, PhysX and Isaac Sim
• NVIDIA Isaac GR00T: NVIDIA open foundation model for humanoid robot reasoning and skills
• OpenVLA: 7B-parameter Vision-Language-Action model for robotic manipulation
• Mobile ALOHA: Bimanual mobile manipulation with 4 robot arms on a wheeled base for teleoperation
• Drake: MIT robotics toolbox for model-based dynamics, optimization, and control system design