Find a question.
Build an experiment.
Six routes through the library, each with a practical outcome. These are editorial starting points, not tutorials we have independently reproduced.
Start in simulation
Inspect a hand model and record one reproducible grasp attempt.
- Pin the model revision and simulator version.
- Check joint names, limits, actuation and object scale.
- Record contact assumptions and one failure.
Capture and retarget motion
Explain how human input maps to the chosen robot embodiment.
- Identify the exact hand, input device and coordinate frames.
- Report latency, dropped frames and human interventions.
- Keep human poses distinct from robot actions.
Build a tactile experiment
Determine whether tactile feedback improves one task over a stated baseline.
- Describe the sensor, calibration and sampling method.
- Compare equivalent trials with and without the feedback.
- Separate rendered touch from measured physical signals.
AnySkinTactile sensing →SparshTactile sensing →TaccelTactile sensing →Tactile Gym 2.0Tactile sensing →
Read the related guide →Choose training data
Select data whose observations, actions and access terms fit the experiment.
- Inspect a sample before planning a full download.
- Keep held-out objects, subjects and episodes out of training.
- Check data, code, model and commercial-use terms separately.
DexYCBHuman motion →ARCTICHuman motion →OakInk2Human motion →DexMimicGenSimulated trajectories →DROIDRobot trajectories →LeRobotLearning tools →
Read the related guide →Document a hand build
Give another builder enough information to reproduce the same revision.
- Record CAD revision, bill of materials and firmware.
- Identify external controllers, power and sensing.
- State what is licensed and which files are actually available.
Measure before you compare
Define a repeatable procedure before reporting a performance claim.
- Use the same objects, starting conditions and success definition.
- Report attempts, failures, resets and human assistance.
- Do not turn unlike tasks or compute timings into a hand ranking.
PoMDARBenchmark →Bench2DexBenchmark →NIST manipulation measurement researchBenchmark →YCB protocols and benchmarksBenchmark →RobotPerfBenchmark →Manufacturing Objects & Assemblies Dataset v2Object geometry →
Read the related guide →Start with the details that matter.
Copy or download a template, fill in what you know, and mark unknowns clearly. Published contributions are public; exclude credentials, personal data and confidential material.
Reproducible experiment
# Experiment title ## Question and claim What are you testing? Which result would support or challenge the claim? ## Setup Hand manufacturer / model / revision: Arm, sensors, controller and external hardware: Firmware, SDK, operating system and dependency versions: Repository URL and exact commit / release: Simulation or physical hardware: Control mode (teleoperated / scripted / autonomous / mixed): ## Procedure Objects and geometry / mass / material: Starting poses and environmental conditions: Step-by-step procedure and reset rule: Success definition, timeout and failure categories (decide before running): Planned trials and variation / randomization: ## Results Attempts / successes / failures (include all attempts): Human interventions and excluded trials with reasons: Timing measurement, units and summary statistics: Raw logs and unedited evidence (where permitted): Baseline and what changed: ## Reproduce it Installation / configuration / command: Expected output and known failures: Limitations and conditions not tested: Code, data, media licenses and original authors: My affiliation / funding / equipment supplied:
Dataset reference
# Dataset name Original authors / institution: Original source and citation: Release / version and retrieval date: ## Scope Human observations, robot actions or simulation: Exact hand / arm / sensor embodiment: Tasks, objects and environments: Sample and episode counts, with units and source: Observations and action schema, frames, units and timestamps: ## Access and reuse Original download / access request URL: Data license; code and model licenses separately: Commercial use / redistribution / consent restrictions: ## Evaluation Train / validation / test split and leakage controls: Missing data, calibration, exclusions and known biases: Useful tasks and unsupported conclusions: One sample inspection / validation procedure: My affiliation and reason for suggesting this resource:
Reference correction
Reference URL: Field or claim to correct: Current value: Proposed value: Exact model / revision / release: Original source URL and publication date: Why this source supports the change: My affiliation (if any):