1. Name the person and the task
Who is the technology for, and what are they trying to do? Be specific about the movement or interaction. “Recognising a comfortable shoulder gesture to select a command” is easier to investigate than “using AI to improve rehabilitation”.
2. Make the question testable
Ask what you need to learn, not only what you want to build. A question might concern reliability across sessions, the effect of sensor placement, a control strategy or the clarity of feedback.
Keep a distinction between an engineering result and a claim about benefit to users. A classifier score alone does not establish that an interface is useful in everyday life.
3. Agree what evidence is available
Identify whether you can use simulation, an existing approved dataset or a practical experiment. Confirm equipment, access, permissions and supervision before making recruitment or data collection central to the project.
Participant involvement, clinical settings and identifiable data require the appropriate institutional processes. Do not collect data before these are agreed.
4. Define a baseline and a fair comparison
Start with a simple reference method. Decide what will count as improvement and how the comparison will be made. Think about test conditions, variation between users or sessions, and likely failure cases.
Useful outcomes may include a carefully justified negative result or an explanation of where a method fails.
5. Make the work reproducible
Document data provenance, preprocessing, assumptions and evaluation choices. Keep code and configuration organised so that someone can understand how a result was produced. Keep private data out of public repositories.
Bring a short project brief
- Question: What do you want to find out?
- Context: Who is the user and what is the task?
- Evidence: What data, tools or equipment could you use?
- Evaluation: What will you compare, and how?
- Constraints: What are your dates, skills and access requirements?