Before you begin
Starting knowledge
- Understand supervised learning and a basic imitation-learning pipeline.
- Be able to inspect dataset splits and robot action conventions.
Equipment & environment
Begin with published model documentation and offline or simulation evaluation; avoid uncontrolled hardware execution.
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
Map model assumptions to the task
Decide whether a model output can be meaningfully evaluated on your setup.
Learn in this order
- Image preprocessing, language inputs and action formats
- Training distributions and embodiment coverage
- Model weights, licenses and compute requirements
Put it into practice
- Compare the documented inputs and outputs of two candidate models.
- Trace one dataset episode into the format required by a chosen baseline.
A model/task compatibility matrix and one validated sample.
Ready for the next step when…
Resources for this stage
Study a vision-language-action model, its training assumptions and intended deployment interfaces.
Explore a compact VLA learning resource; verify the policy and hardware assumptions in its documentation.
Examine multi-embodiment demonstration data, metadata, licenses and action conventions.
Intermediate
Create a matched evaluation harness
Keep tasks, resets and metrics constant across model choices.
Learn in this order
- Task splits, checkpoints and deterministic setup
- Latency, intervention and completion metrics
- Prompt changes and failure categorization
Put it into practice
- Evaluate a small fixed task set using documented benchmark protocols.
- Run an ablation on one input condition and keep the rest unchanged.
A versioned evaluation harness and baseline comparison.
Ready for the next step when…
Resources for this stage
Investigate a manipulation benchmark and preserve its task definitions, splits and evaluation conditions.
Follow robot-data recording, dataset, policy and evaluation workflows for supported hardware and software.
Build repeatable manipulation tasks in a simulated benchmark environment.
Professional
Audit robustness and deployment boundaries
Produce an evidence-based statement of model capability.
Learn in this order
- Generalization versus memorized scene conditions
- Compute and control-loop constraints
- Real-robot transfer and supervised validation
Put it into practice
- Test a declared set of object, language or visual shifts.
- Write a deployment decision with unsupported conditions and escalation rules.
A robustness report, resource budget and deployment-readiness decision.
Ready for the next step when…
Resources for this stage
Explore how manipulation policies are evaluated; separate benchmark evidence from deployment claims.
Study a manipulation data-collection workflow and its embodiment and environment assumptions.
Explore robotics simulation and synthetic scenes; check hardware requirements and release compatibility.
Common pitfalls
- Comparing headline scores from incompatible evaluation protocols.
- Treating language fluency as evidence of reliable physical action.
Where to go next
Connect the evaluated model to a supervised manipulation project only after compatibility and latency gates are met.
Explore other research paths →