A care robot does not stop needing care when its launch ends. Batteries, sensors, wheels, harnesses, software, data, networks, and human knowledge deteriorate or change. If nobody receives time, authority, and money to sustain them, a technological promise becomes risk for the person depending on it. Maintenance is silent justice because it preserves safety and autonomy every day, including after photographs, demonstrations, and exceptional funding have disappeared.

Launch is not the end of design#

A launch offers an orderly scene: the device is charged, calibrated, and accompanied by specialists. Real service life follows with shifts, dust, impacts, updates, irregular connectivity, and changing people. Treating installation as project completion separates technology from the setting that sustains it. A useful robot needs operational design throughout its cycle, including withdrawal.

A launch offers an orderly scene: the device is charged, calibrated, and accompanied by specialists.

Public attention rewards novelty. Acquisition may have a dedicated budget while cleaning, inspection, and training compete with ordinary tasks. This asymmetry creates a maintenance debt that does not appear immediately. Small signs accumulate: shorter battery life, sensors requiring restarts, repeated alerts, and outdated manuals. By the time failure becomes visible, someone needing the service has already carried uncertainty and burden.

Justice resides in continuity. A person should not lose mobility, medication, navigation, or communication because a grant ended or a supplier changed. Services retain human alternatives and critical spares. Plans state which functions are essential, how long they may be unavailable, and which response follows. Actual dependency, not device prestige, determines priority.

Inventorying more than machines#

A responsible inventory records more than model and serial number. It includes purpose, location, population, software version, sensors, consumables, licences, network, accessories, data, owner, supplier, schedule, alternative, and support end date. Required competencies are identified too. Without this information, a failure circulates among departments while an affected person searches for a response.

Dependencies are often invisible. A transfer robot needs the correct sling, space, charge, calibration, and two professionals in particular situations. A social assistant relies on profiles, speakers, connectivity, and human response. A logistics system requires compatible doors, maps, and lifts. Maintaining the robot without maintaining its chain preserves the appearance of availability while losing the function.

Documentation must be accessible and usable under pressure. Long manufacturer instructions become concise procedures for inspection, stopping, and escalation without removing necessary technical detail. Contacts and hours remain current. Copies exist where teams work rather than only in an administrative folder. Simulation can reveal that a written protocol does not match reality.

Physical, digital, and relational maintenance#

The physical dimension covers cleaning, wear, calibration, batteries, mobility, grips, signals, and stopping. Frequency reflects risk and use rather than only a general recommendation. Complete functions are tested with realistic loads and spaces. A green status light does not prove safe movement or stopping. Replacement parts and tools must be available within timeframes compatible with care.

The digital dimension includes software, models, settings, permissions, logs, cybersecurity, and interoperability. An update may correct a vulnerability while changing voice, speed, or accessibility. It is therefore tested beforehand, documented, and supported by a rollback plan. Significant changes are communicated and may require renewed choice. Keeping an obsolete version is not neutral either; it can increase security risks or lose support.

The relational dimension sustains understanding and trust. People need to know whether the robot changed, which limitations exist, and where to seek help. Staff need time to explain and listen. If failure creates fear, a technical restart is insufficient. Repair acknowledges experience, offers an alternative, and permits a decision about resuming. Care cannot be reduced to hardware status.

Curiosity that detects deterioration before harm#

Curiosity turns ordinary signals into questions. Why are rescues increasing on one route? Why do some people repeat commands? Why does one shift use the robot less? The answer is not automatically resistance or incompetence. Wear, noise, lighting, updates, training, load, and design are examined. A small pattern may anticipate a serious incident.

Controls combine scheduled inspection, limited telemetry, observation, and reports. Monitoring does not justify continuous surveillance of users or workers. Necessary data is collected with defined purpose, access, and retention. People know what is recorded. Predictive maintenance may help, but its models also need validation and do not replace physical review.

Near misses provide knowledge. Delayed stopping, an ignored warning, or a manual intervention without harm is documented proportionately. Reporters receive a response and protection. An initial rise in reports may indicate greater trust. The organisation analyses recurrence and severity rather than rewarding an empty log. Silence does not prove safety.

Responsibility with names, time, and budgets#

Every task needs a primary owner and backup. Engineering may address technical failures, but care operations decide continuity; procurement manages suppliers but cannot replace oversight; privacy reviews data while operations apply permissions. A matrix clarifies who authorises stopping, repair, and restart. Users should not coordinate departments to recover an essential function.

Maintenance time belongs in workforce planning. Charging, cleaning, testing, and documenting are not favours or invisible activities at the end of a shift. Training receives recognition and staffing is adjusted. Where relatives configure a domestic robot, services do not presume they will continue. Professional support is supplied and a refusal of that role respected.

Budgets cover total cost: licences, parts, consumables, connectivity, support, audit, accessibility, cybersecurity, training, alternatives, and withdrawal. The cheapest acquisition can become costly if irreparable or tied to one supplier. Contracts cover service levels, change notification, export, data deletion, and end of support. The organisation retains stopping capability.

Case 1: a domestic exoskeleton and its battery#

A person uses an exoskeleton for short journeys at home. Over time, battery endurance falls and the indicator changes unpredictably. The service considers the device within its annual inspection period. The user plans movement anxiously and reduces activity to avoid becoming immobilised. Her partner begins accompanying every journey. Promised autonomy becomes silent dependency before an official breakdown.

The programme introduces capacity monitoring, a backup battery, replacement thresholds, and an accessible channel. The device is assessed in its actual environment, while the user learns signals without inheriting technical responsibility. Her partner stops being a compulsory control. Chosen support is offered during replacement. Indicators include perceived reliability and avoided activities, not only powered hours. Maintenance restores freedom before harm occurs.

Case 2: a disinfection robot with degraded sensors#

A hospital uses a disinfection robot. After several months, dust and impacts affect a sensor, and the device stops cycles frequently. Staff restart it to meet schedules because each stop lowers productivity. The dashboard records initiated cycles rather than coverage or interventions. The machine appears available while quality and safety depend on human shortcuts.

Management suspends use, validates coverage, and reviews potentially affected areas. Cleaning and calibration frequency changes, the sensor receives protection, and justified stops are separated from performance. Cycles record completion and configuration. Staff gain authority to stop without penalty. A supplier provides support, but the hospital retains independent verification. Repair addresses exposure and corrects the metric that encouraged shortcuts.

Case 3: a social robot at the end of support#

A residential service uses a social robot with profiles and reminders. The manufacturer announces closure of its cloud service. Management budgeted for purchase but not migration. Residents fear losing routines and data, while staff do not know which functions require the cloud. Continuing without updates would increase risk, but abrupt removal would disrupt support and connections attributed to the machine.

The centre inventories functions, safely exports necessary information, deletes remaining data, and explains limitations. Chosen human and technological alternatives are offered. Withdrawal is gradual and emotionally supported without claiming the robot will feel loss. Future contracts require portability and exit planning. The experience shows that end of life belongs to maintenance and continuity of care cannot depend on commercial continuity.

Cooperation between technical and lived knowledge#

Manufacturers know components; operators know flows; people receiving care know effects; accessibility, safety, and privacy specialists see other risks. Reviews combine these perspectives. No voice is decorative. Users and workers participate in priorities and are paid where evaluation becomes work. Final accountability stays with the institution.

Cross-training avoids dependency on one person. Several shifts can inspect, stop, and refer; technical teams understand care consequences; management knows limitations. Suppliers explain changes in usable language. Rotation, holidays, or absence do not remove all knowledge. Simulations cover network, power, sensor, and unavailable-support scenarios.

Governance committees review incidents, workload, refusal, inequality, and updates. They have authority and schedules. Information is accessible and includes cases alongside averages. Greater failures in one group trigger investigation. Cooperation turns an individual incident into shared learning without exposing identities.

Indicators of continuity and justice#

Availability should measure actual function, not a powered device. Downtime, affected tasks, interventions, recovery, repeated failure, alternative quality, and completed maintenance are recorded. Burdens for users, families, and staff are included. An overall percentage may hide that the sole essential function fails at night. Data is separated by context with protection.

Autonomy is measured as well: possible activities, need for permission, confidentiality, choice, and confidence. Rapid repair is insufficient if the person stops using the robot. Accessible interviews complement logs. Satisfaction is not demanded, and emotion is not reduced to a simplistic score. An account helps explain why a figure changed.

Maintenance actions receive later verification. Closing a work order does not prove resolution. The task is repeated, observed over time, and discussed with those performing it. Controls can create new barriers. Each change keeps traceability and a review date. Humility accepts that a solution may require another correction.

Repairing failures of care and system#

When deterioration causes harm, the first response attends to the person, stops risk, and offers continuity. Information is clear and records are corrected. Expenses, time, opportunity, and burden may require compensation. An apology accepts responsibility without blaming someone for not reporting earlier. The person chooses their participation in investigation.

System repair changes frequency, design, staffing, budget, contract, or purpose. Every action has an owner and deadline. If control is insufficient, the function is limited or withdrawn. Prior investment does not justify continued risk. Aggregate learning is shared so other centres do not repeat it. Maintenance also means knowing when to stop use.

An accessible channel permits reopening if the solution fails. Progress communication sustains trust without promising perfection. Complete repair cannot always return what was lost, but it can acknowledge that loss and reduce repetition. Silent justice appears when change remains after case closure.

Institutional sources for the life cycle#

The World Health Organization provides guidance on ethics and governance of artificial intelligence for health. The NIST AI Risk Management Framework structures governance, identification, measurement, and continuous management. Both support attention extending beyond deployment.

The United States Food and Drug Administration provides institutional information about medical-device cybersecurity. The International Labour Organization supplies resources concerning the care economy and decent work. These sources need applicable local rules, manufacturer instructions, technical evaluation, and user experience.

A useful robot tomorrow, not only today#

Maintenance reveals which values survive after novelty disappears. Curiosity detects signals, responsibility assigns resources, and care protects continuity. A robot that knows its place does not pretend to work alone: it communicates limits, accepts stopping, and depends upon acknowledged human infrastructure. Technology becomes more reliable when it abandons the fiction of total autonomy.

Everyday justice is built through timely battery replacement, calibrated sensors, secure software, trained teams, protected data, and available alternatives. These tasks rarely produce a ceremony, yet they sustain freedom and safety. Maintenance is not preserving a machine out of loyalty to investment. It preserves care while the tool remains beneficial and knows how to transform or withdraw it when that ceases to be true.