Define The Human-Proximity Problem
Humanoid robots promise flexible automation in spaces designed for people, but shared space creates a different safety problem from fenced industrial automation. A robot that can move through aisles, handle objects, work near workers, and adapt to changing tasks must prove more than task performance. It must prove that people remain protected when normal work becomes messy.
A human-proximity safety case is the evidence file for that proof. It records where the robot may operate, how it detects and responds to people, what tasks are allowed, what conditions require stopping, who trains workers, how incidents are reviewed, and what evidence shows the controls still work after changes.
Why Digit 5 Is A Useful Signal
The Daily AI Newsflash highlighted coverage of Agility's Digit 5 and its push toward safer operation near people. Agility's public materials describe Digit as a humanoid robot with extensive production experience and say Digit 5 safety features are being developed for cooperatively safe work at scale, while also noting that safety features do not eliminate all operational risk.
That last point matters. Better sensing, stopping, posture control, and fleet software may expand where humanoids can operate, but they do not remove the employer's responsibility to design safe work. The practical question for a business is not whether a vendor says the robot can work near people. It is whether this site, task, traffic pattern, and supervision model have a defensible safety case.
Count The Cost Of Assumed Safety
Assumed safety is expensive because one incident can stop a deployment, harm a worker, trigger regulatory review, damage trust, and force redesign after equipment has already been purchased. Even near misses matter. Workers who do not trust a robot will route around it, slow down, or create informal rules that management cannot see.
A simple estimate starts with downtime and investigation. If a humanoid line pilot stops for three days after a near miss and twelve people lose two hours per day to rescheduling, the visible lost time is seventy-two staff hours before engineering review, vendor support, retraining, and morale are counted. Prevention is cheaper when the safety case exists before the pilot starts.
Diagnose Site Readiness
Start with the physical workspace. Where do people walk, carry loads, turn corners, open doors, stage pallets, clean spills, and move equipment? Where does the robot need to pass, reach, lift, set down, charge, or wait? Human-proximity safety depends on these ordinary details.
Warning signs include unclear zones, mixed pedestrian and robot traffic, tasks that change without review, no process for worker feedback, no near-miss log, and no defined authority to stop the deployment. If workers cannot explain what the robot is allowed to do and what they should do when it behaves unexpectedly, the site is not ready for shared operation.
Choose The Safety Case Evidence
The evidence file should include task description, payload limits, speed and stop behavior, sensing modes, exclusion zones, shared zones, worker approach rules, emergency stop locations, training records, maintenance checks, software version, incident records, near-miss reviews, and change approvals. Each item supports a specific safety claim.
The file should also name the limits of the case. A robot approved to move empty totes in a marked aisle is not automatically approved to work in a crowded receiving dock, handle sharp objects, operate during cleaning, or pass through a visitor area. Humanoid flexibility is useful only when each new use is reviewed before it becomes routine.
Build The Human-Proximity Safety Case
A practical safety case begins with a narrow task and a mapped workcell or route. The team defines allowed interactions, tests stop behavior, observes worker movement, records failure cases, and assigns an owner for daily checks. Vendor documentation is input, but the site-specific evidence must come from the actual workplace.
The case should be updated whenever task, layout, payload, speed, software, staffing, shift pattern, or nearby equipment changes. It should also be readable by operations leaders, safety staff, supervisors, and workers. A safety case that only engineers understand will not govern the shared space where people actually work.
Worked Example: Tote Movement Near Workers
Imagine a warehouse pilot where a humanoid moves empty totes from a packing area to a staging rack. The safety case defines a fixed route, maximum tote weight, worker crossing points, robot waiting zones, stop distance, alert behavior, and supervisor response. Workers are trained to use the route markings and report any unexpected approach.
During the pilot, a temporary pallet blocks the route. The robot stops and waits, but workers begin carrying totes around it. The near-miss review shows that the route-blocking procedure was incomplete. The safety case is updated with a blocked-route rule, a supervisor alert, and a requirement that temporary staging never enter the robot path.
Measure Shared-Space Safety
Useful measures include near misses, unexpected stops, route blockages, worker interventions, emergency-stop uses, task exceptions, training completion, maintenance findings, and time from safety issue to corrective action. These measures should be reviewed alongside productivity, because throughput without safety evidence is not a successful deployment.
The quality measure is whether evidence changes behavior. If near-miss reports do not alter routes, rules, speed, training, or task design, the safety case is only paperwork. A living case helps the business decide when to expand, when to slow down, and when a use case is not ready for human-proximity operation.
Start With One Shared-Space Task
Before buying or expanding humanoid automation, select one task and write the first safety case. Map the space, state the safety claims, gather vendor evidence, test the site controls, train the workers, and define the stop rule. Keep the first case narrow enough to verify.
If the first case is hard to write, that is useful information. It may show that the task is too variable, the layout is too crowded, or the organization has not decided who owns robot safety after installation. Humanoid robots may become practical tools, but shared workspace safety has to be proven one operating case at a time.
Sources And Methodology
This article uses the Daily AI Newsflash item on Agility's Digit 5 and humanoid safety as the news trigger. It also references Agility Robotics' public information on Digit, fleet operation, and safety notes, ISO's ISO/TS 15066 collaborative robot specification, and the Association for Advancing Automation's robot safety standards resources.
The human-proximity safety case is SynHy analysis for businesses evaluating humanoid or mobile robotic deployments. It is not a safety certification, engineering approval, legal conclusion, or claim that any named robot is unsafe. A real deployment should involve qualified safety professionals, the robot supplier, workers, supervisors, and applicable regulatory requirements.