Design begins by defining the task and the people who share it
Aliases: HRI problem framing · task ecology · work-system analysis
What it is
Task–human–robot framing specifies the work a robot is meant to accomplish, the people who share that work, and their dependencies before selecting form or features. The same platform used in a warehouse, ward, and home faces different success criteria, tolerable failures, explanation needs, and opportunities for intervention. “User” may include operator, recipient, maintainer, and involuntarily exposed bystander.
Why it happens
Robot capability is not context-free. The task determines action sequences and failure consequences; people determine available instructions, physical capacity, and responsibility. Together they determine what must be sensed and when help is required. Technology-first framing can substitute demonstration success for work success and leave maintenance, handoff, and exception recovery as invisible labour performed by people on site.
Studying it
Contextual interviews, task analysis, shadowing, and critical-incident elicitation can map goals, prerequisites, information flows, collaborators, variability, and recovery. The unit of analysis should be the complete work loop rather than only moments of direct conversation with the robot. Sampling across roles and shifts matters: interviewing only buyers or primary operators omits those who absorb noise, waiting, interruption, or risk.
Where it stops holding
Task definition does not imply a permanently fixed workflow. Goals change in emergency, care, and public-service settings, and premature narrowing can hide necessary adaptation. Exploratory research may begin with a capability prototype, but deployment still requires translating “what the technology can do” into whose task improves and to whom the remaining cost is transferred.
Applying it
- For each goal, state triggers, completion criteria, failure consequences, and collaborating roles rather than using broad verbs such as “assist.”
- Map the nominal flow and at least several exception loops, assigning information, action authority, and recovery responsibility at every step.
- Rehearse the end-to-end task on site. Measure nominal time, manual recovery time, interrupted roles, and near misses. If improvement depends on uncounted human monitoring, reallocate the work.