Visual Dexterity
Improbable-AI
Visual learning for in-hand object reorientation with a D’Claw robot.
What you can explore
Study policies that use visual observations to handle different object shapes.
Before you use it
The experimental robot is D’Claw, not an anthropomorphic five-finger hand. The released code depends on legacy Isaac Gym.
Original contributors
Improbable AI / Visual Dexterity research team
License / access: Code: MIT · data, models and dependencies may have separate terms
Keep exploring
HORA
In-hand object rotation using reinforcement learning and rapid motor adaptation.
DexPoint
Point-cloud reinforcement learning for generalizable dexterous manipulation.
Bi-DexHands
A collection of bimanual dexterous manipulation tasks for reinforcement-learning experiments.
Editorial reference · checked 2026-09-22