Before you begin
Starting knowledge
- Comfort with Python, data arrays and supervised-learning concepts.
- Understand the robot and camera calibration procedure.
Equipment & environment
Use an offline dataset or simulator first. Hardware demonstrations need a supervised recording space and documented reset procedure.
No fixed completion time: use the evidence gates below to decide when to advance. Resource links open the botstest directory entry; check its publisher documentation for the exact version and configuration.
Beginner
Inspect the data before training
Understand how observations, actions and episodes are represented.
Learn in this order
- Episode boundaries, timestamps and units
- Action representation and embodiment
- Dataset provenance and permitted use
Put it into practice
- Load a small demonstration dataset and inspect frames and actions.
- Write a schema note and identify missing or misaligned observations.
A dataset inspection notebook and schema/permission note.
Ready for the next step when…
Resources for this stage
Follow robot-data recording, dataset, policy and evaluation workflows for supported hardware and software.
Inspect a manipulation dataset and its collection conditions before designing training and evaluation splits.
Examine multi-embodiment demonstration data, metadata, licenses and action conventions.
Intermediate
Record and train a bounded task
Build a reliable collection-to-training loop for one task.
Learn in this order
- Teleoperation calibration and operator consistency
- Held-out episodes and object leakage
- Baseline training configuration and checkpoint selection
Put it into practice
- Collect or curate a small fixed-task dataset and train a supported baseline.
- Compare two data-quality conditions using the same evaluation setup.
A versioned dataset subset, baseline checkpoint and quality comparison.
Ready for the next step when…
Resources for this stage
Explore a small teleoperation arm; confirm assembly, calibration and exact hardware revision support.
Follow robot-data recording, dataset, policy and evaluation workflows for supported hardware and software.
Study offline demonstration-learning workflows and controlled policy comparisons.
Professional
Evaluate generalization and release the workflow
State where the learned policy works and where it needs intervention.
Learn in this order
- Shifted objects, scenes and start states
- Intervention accounting and recovery
- Cross-embodiment and hardware-transfer limitations
Put it into practice
- Run a predeclared evaluation matrix with repeatable resets.
- Release collection, processing and evaluation instructions with an honest failure report.
A complete demonstration-to-evaluation report and reproducible artifacts.
Ready for the next step when…
Resources for this stage
Study the original bimanual research design. Legacy commercial ALOHA hardware is end-of-life; verify parts and current alternatives.
Examine handheld demonstration collection and how its calibration and action representation transfer to a target robot.
Study a manipulation data-collection workflow and its embodiment and environment assumptions.
Common pitfalls
- Assuming more demonstrations fix a broken timestamp convention.
- Mixing evaluation scenes into training without disclosure.
Where to go next
Proceed to VLA comparison only when the simpler policy and data-quality baseline are understood.
Explore other research paths →