Extrinsic dexterity
- Category: Planning & manipulation › Manipulation
- Tags: manipulation, contact, in-hand-manipulation, grasping, nonprehensile-manipulation
Manipulating an object held in the hand by using resources outside the hand, such as contact with the environment, gravity or arm motion, to do what the fingers alone cannot.
Definition
Extrinsic dexterity is a robot's ability to reposition or reorient an object in its grasp, or to get it into a graspable pose, by using forces from outside the hand: contact with a table, wall or fixture, gravity, or the inertia produced by accelerating the arm. The term was introduced by Nikhil Chavan-Dafle, Alberto Rodriguez and colleagues in 2014 to describe how a simple gripper with few actuated fingers can still perform in-hand manipulation by pushing the held object against the environment. It contrasts with intrinsic dexterity, which relies on the hand's own actuated fingers.
Typical strategies include pivoting an object about the grasp axis by pressing its edge against a surface, sliding an object between the fingers by pushing it against a wall, pushing a thin object such as a plate or a card against a wall so that its edge lifts off the table and can be grasped, letting gravity swing or slip an object into a new pose, and regrasping by setting an object down and picking it up again differently. People do this constantly, for example sliding a coin to the edge of a table to pick it up.
Planning these actions is hard because they are contact-rich: the outcome depends on which contacts stick or slide, both between object and gripper and between object and environment, and on friction values that are uncertain. Approaches include quasi-static mechanics with friction cones and planning over contact modes, contact-implicit and smoothed-contact planners that search over contact sequences automatically, and learning-based policies trained in simulation. Compliance, either in the gripper or through force or impedance control in the arm, is important in practice because it keeps contact forces bounded while the object moves against the environment.
Why it matters
Most robots in industry use simple parallel-jaw or suction grippers. Extrinsic dexterity extends what such grippers can do without more complex hands, and it describes much of how people actually handle objects.
Key applications
- Picking thin or flat objects from tables and bins
- Reorienting parts for assembly using fixtures
- In-hand regrasping with simple grippers
- Pivoting and repositioning tools
- Logistics bin picking
In practice
A warehouse team needs to pick thin, flat items such as books lying flat in a bin, where the gripper fingers cannot get underneath. Their strategy is to press down on the item with one finger, slide it against the bin wall until its far edge tilts up, then grasp the raised edge. The arm runs in impedance control so that the downward force stays within a set range while sliding. The approach only works if the friction between the finger pad and the item exceeds the friction between the item and the bin floor; otherwise the finger slips over the item. Most failures they see come from this, so they choose a high-friction finger pad and detect slipping from the arm's force readings.
Common misconceptions
- Dexterous manipulation requires a multi-fingered hand — many in-hand manipulations can be done with a parallel gripper by using the environment
- Contact with the environment should always be avoided — collision-free planning treats it as failure, but deliberate, controlled contact is often the simplest route to a new grasp
- Extrinsic dexterity is just pushing — it also covers using gravity, dynamic arm motions, and setting an object down to regrasp it
Who should know this
- Manipulation and grasping researchers
- Warehouse and logistics automation engineers
- Gripper designers
- Assembly automation engineers
Related technologies
- Parallel-Jaw Grippers
- Impedance Control
- Quasi-Static Pushing Models
- Contact-Implicit Planning
- Compliant Fingertips
Related terms
- Dexterous Manipulation
- Contact mode
- Friction Cone
- Gripper
- Impedance Control
- Hybrid Position/Force Control
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
- Drake: MIT robotics toolbox for model-based dynamics, optimization, and control system design
- OpenArm: 7DOF humanoid arm for physical AI research in contact-rich environments