Before you begin
Starting knowledge
- Basic Python and linear algebra.
- Understand camera, base and tool coordinate frames.
Equipment & environment
Start with a supported simulation task. Add real hardware only with lab supervision and a task-specific operating 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
Turn the question into an experiment
Define the manipulation task and what would count as evidence.
Learn in this order
- Initial-state distribution and success definition
- Environment versions and deterministic resets
- Observation, action and control-rate conventions
Put it into practice
- Reproduce a simple manipulation baseline in simulation.
- Write the trial reset and record a successful and failed episode.
A baseline protocol and a small inspectable episode set.
Ready for the next step when…
Resources for this stage
Build repeatable manipulation tasks in a simulated benchmark environment.
Model physical interactions and test control or learning behavior under explicit simulation assumptions.
Study the mechanics, coordinate frames, kinematics and control foundations relevant to the exercise.
Intermediate
Control variation and compare methods
Change one experimental factor without changing the evaluation task.
Learn in this order
- Train/evaluation separation and trial budgets
- Ablations and parameter sweeps
- Camera and robot calibration assumptions
Put it into practice
- Compare two methods with the same held-out task setup.
- Log interventions, resets and latency alongside success.
A comparison notebook, configuration snapshots and failure taxonomy.
Ready for the next step when…
Resources for this stage
Study offline demonstration-learning workflows and controlled policy comparisons.
Inspect a manipulation dataset and its collection conditions before designing training and evaluation splits.
Visualize spatial observations and compare them with model and trajectory outputs.
Professional
Produce a publishable experimental artifact
Make the claims traceable to data, code and operating conditions.
Learn in this order
- Uncertainty, repeated runs and release metadata
- Simulation-to-real limitations
- Dataset permissions and reproducible environment capture
Put it into practice
- Prepare a clean rerun package with exact dependencies.
- If transferring to hardware, document the differences and supervised acceptance tests.
An experiment package, limitations section and reproducibility checklist.
Ready for the next step when…
Resources for this stage
Investigate a force-sensitive manipulation platform and confirm the exact research interfaces and configuration.
Compare an arm research platform against the task, tooling and software interfaces you need.
Explore how manipulation policies are evaluated; separate benchmark evidence from deployment claims.
Common pitfalls
- Changing object positions or trial counts between methods.
- Reporting successful videos without total attempts.
Where to go next
Extend into imitation learning, VLA evaluation or contact sensing while keeping the same reproducibility contract.
Explore other research paths →