Recruiting and Briefing Demonstrators for Commissioned Collection

Why the demonstrator is a specification item

Teams spend weeks defining camera rigs and label schemas, then leave "who performs the task" to whoever the site offers. That choice shapes every downstream property: how fast and fluently the task is done, how many distinct techniques appear, whether the dataset covers left- and right-handed workers, and which people carry rights to be told what happens to their image.

The skill question has its own page, expert vs novice demonstrations. This page covers recruitment and briefing: where demonstrators come from, what they need to know, and how to avoid consent that is formal but not real.

Demonstrator sources

Combine sources where the task calls for it. Large research efforts use many participants: Ego4D reports 931 unique camera wearers over 3,670 hours, in 74 locations and 9 countries, with consenting participants and de-identification procedures (Grauman et al., 2021). A workplace program will usually be smaller, but the lesson transfers: diversity of wearers and locations is a design choice made through recruitment.

Define what you need before recruiting

• Skill range. Experts only, a mix, or labeled tiers? Record the tier for each take.
• Variation in technique. Several operators doing the same step differently is valuable for generalization, and harmful if it is unlabeled noise.
• Handedness, height, build, and glove or PPE use where the robot should cope with them.
• Number of demonstrators versus takes per demonstrator. A few people repeating many takes gives volume but less diversity; the reverse costs more per take.
• Experience with wearable rigs. First-day wearers behave differently from habituated ones.
• Holdouts. If you will evaluate on unseen operators, reserve whole demonstrators; see demonstration dataset splits.

Document these in a demonstration coverage plan so collection teams recruit to a spec.

Voluntary participation in an employment setting

When demonstrators are employees, consent has an extra complication: an employee may feel unable to refuse. Regulators address this in monitoring guidance. The UK Information Commissioner's Office guidance on monitoring workers, published October 2023, recommends a data protection impact assessment for monitoring likely to result in a high risk to workers' rights and freedoms (ICO, 2023). The EU's GDPR lets Member States set more specific rules for processing employees' personal data in the employment context (Art. 88(1)), and lists biometric data processed to uniquely identify a person among special categories (Art. 9(1)) (GDPR, 2016). Those are references from particular jurisdictions. This is not legal advice, and the law that applies depends on where the demonstrators and the recording are.

Practical diligence points for any jurisdiction:

• Participation is optional, with no consequence for declining and a way to withdraw.
• The person asking is not the person who evaluates their work.
• Pay or time off is stated, and does not depend on speed or output.
• The consent form states in plain language what is recorded, who can see it, how long it is kept, and that the data may be used to train AI systems and be licensed to third parties.
• The form covers each sensor type (video, depth, audio, hand tracking, IMU) rather than only "recording".

See worker consent and identifiable people for the procurement view.

Briefing content

A briefing turns recruitment into usable data. Cover:

1. Purpose in plain terms: demonstrations for training or evaluating robot systems.
2. Rights: voluntary, can stop, can see how their data is described, who to contact.
3. The task script: steps to perform, which are fixed and which are free, and how to signal start and end of each step.
4. Natural pace: do the task as you normally would, without hurrying or exaggerating. See working pace in demonstrations.
5. Equipment: how the rig is worn, what to avoid touching, what to do if it slips.
6. Mistakes and recoveries: say whether to continue after a slip. Most buyers want these labeled, not hidden; see failure and recovery demonstrations.
7. Privacy: no-record zones, what to do if a bystander enters.
8. Safety: the site's rules apply, and safety overrides the script.

Site-side rules for notice, stop conditions and handover are in recording protocols for live work sites. Let demonstrators rehearse with the rig before recording, and test the whole briefing in a pilot collection first. The first minutes are typically unrepresentative.

Limitations

Careful recruitment improves a dataset but cannot fix everything:

• Small pools limit diversity. A single site may offer a few willing people with similar training. Diversity of technique, body size and handedness may need several sites or recruitment routes.
• Being recorded changes behavior. Demonstrators may slow down, tidy their method or skip shortcuts they normally take. Habituation sessions help but do not remove the effect; compare pace and method against unrecorded observation where you can.
• Skill labels are judgments. Tiers assigned by supervisors or by years of experience are proxies. Record how a tier was assigned so later users can weigh it.
• Consent can narrow later. People can withdraw, and their takes may need to be removed after delivery. Plan splits and version control so that removal does not break evaluation sets.
• Human bodies are not robot bodies. Recruiting people whose height or reach resembles a humanoid narrows the gap only partly; kinematics, hand degrees of freedom and strength still differ, and human demonstrations alone do not supply robot actions.

Illustrative demonstrator record

# Illustrative only - synthetic values, pseudonymous IDs
demonstrator_id: operator_A12
source: host_employee | agency_worker | recruited | trainer
skill_tier: novice | experienced | expert   # assigned by buyer-agreed rule
experience_years_task: 4
handedness: right
ppe_in_use: [nitrile gloves, safety glasses]
rig_familiarity_sessions: 2
consent:
  form_version: v2
  scopes: [video, hand_tracking, imu, training_use, third_party_license]
  withdrawal_contact: provided
  signed_date: 2026-05-02        # synthetic
sessions: [site_017_s01, site_017_s02]

Keep identity linkage outside the delivered dataset. Deliver pseudonymous IDs; the link to a name stays with the collection team under the retention terms in the license.

Questions to ask collection teams

1. Where will demonstrators come from, and how are they invited?
2. How do you make participation voluntary for employees of the host site?
3. How do you assess and label skill?
4. What does the consent form cover, sensor by sensor, and can I review it?
5. How many distinct demonstrators and how many takes per person are planned?
6. What is the rehearsal and briefing procedure?
7. Who sees recordings of demonstrators, and can they withdraw after recording?
8. How is payment structured, and does it affect pace or output?
9. What demonstrator metadata is delivered, and in what form?

Discuss your requirements

If you need new demonstrations performed by people with particular skills or backgrounds, describe the task, skill range and permitted uses to SourceX. If no existing asset fits, SourceX can scope a commissioned collection of new recordings; SourceX's role is sourcing, rights review, licensing and delivery coordination, with specialist capture partners engaged case by case. A request does not guarantee a match. See the site's physical-world path for new recordings of hands-on work, the humanoid training data guide and the workplace task demonstration hub.

Describe the workplace demonstration data your robotics project needs

Sources

• Ego4D: Around the World in 3,000 Hours of Egocentric Video — number of wearers and locations; consent and de-identification.
• Employment practices and data protection: monitoring workers — UK guidance on monitoring workers.
• Regulation (EU) 2016/679 (General Data Protection Regulation) — employment-context rules and special categories.