DEXTEROUS MANIPULATION
The world of dexterous hands.
Compare hardware. Discover applications. Share what works. An open reference and community for dexterous manipulation.
Learn
From hand mechanics to the algorithms that make manipulation possible.
Dexterity Intelligence
Understand VLA and world models. Trace their actions, inspect the evidence and design your own experiment.
Start with a real hand: Prima 1
Prima 1 is the worked example across our lessons. Choose a task, check its interfaces and download your first experiment brief. Referenced research keeps its original hardware and scope.
Explore the Prima 1 learning path →Build your foundations
Dexterity Foundations
Understand grasping, joints and independent motion. Start here if robotic hands are new to you.
Start learningTouch & Contact
Explore what happens where fingers meet an object, and why sensing needs a feedback loop.
Start learningControl & Teleoperation
Follow a command from a human to a robot. Learn why retargeting is more than copying angles.
Start learningStart in Simulation
Choose a documented model, check its limits and design your first repeatable experiment.
Start learningData for Dexterity
Tell human motion from robot actions. Choose data that fits your embodiment and learning task.
Start learningEvidence & Evaluation
Look beyond a successful clip. Count attempts, record interventions and make results reproducible.
Start learningDexterity Intelligence
Understand VLA and world models, trace their actions and design a defensible experiment.
Start learningCompare & Choose
Compare motion, sensing and software, then read prices and quotes on the same basis.
Start learningPut it into practice
Four practical guides with worked examples, a task to complete and a knowledge check.
Try a concept in your browser
No installation or account required.
ALGORITHMS & MODELS
Learning & Control
Explore how a dexterous hand perceives, predicts and acts. Choose a research direction.
New to robot learning? Learn the hand, understand its controller, then define a test before choosing a model.
Imitation & diffusion
Learn actions from demonstrations.
Imitation learning trains on examples of observations and actions. Diffusion policies are one way to generate action sequences. Demonstration coverage and the robot action convention matter as much as the model choice.
Start with: Synchronized observation/action episodes and behavior-cloning basics.
- InputDemonstration observations + actions
- MethodBehavior cloning / diffusion training
- OutputPolicy that proposes actions
Human hand poses, robot joint commands and generated video are different training targets.
How to evaluate it
Held-out task success, failure types and data coverage; keep full evaluation episodes out of training.
A first experiment to try
Inspect one episode, document units and timestamps, and keep whole held-out episodes out of training. Start with a reproducible baseline before changing the architecture.
Related practical lesson →An editorial learning map, not a model ranking. Follow each project for its original paper, code, release scope and limitations. Models run through their original projects; Dexhands does not host model inference.
Keep exploring
Your learning progress →Projects with setup guides →Explore dataset formats →Research collections →Tools & documentation →Research labs →Events & deadlinesOriginal Dexhands lessons with links to the underlying research. Browser activities teach concepts; they do not test a real hand or replace a simulator.