What you can do
Adapt to your task
A workflow that collects demonstrations into training data, trains further, then re-scores on the same test. Whether adaptation is included, and its scope, are agreed in a joint pilot test.
1
Collect demonstrations
2
Convert to training data
3
Further training
4
Re-score on the same test
5
Next adaptation from failure causes
What you get
Same-test results before and after adaptation · failure causes
Limitations
Adaptation is joint work with a scope set by agreement. Performance after adaptation is not promised in advance.
Features used
Test subject
Task adaptation workflow
Collects demonstrations into training data, trains further, then re-scores on the same test. Scope is set by agreement.
Run · connection
Training data export
Exports test recordings (robot cameras, state, actions) as training data in LeRobot and RLDS formats.
Test subject
Learned policy execution · reinforcement learning
Attaches trained neural network policies to robots for testing, and trains policies with reinforcement learning.
Scenario
Repeat · comparison experiments
Groups runs for repetition, compares runs with each other and checks hypotheses registered in advance.
Domains
In which domains
Get started
How to use
Other things you can do