Name, definition, and system boundary
A moxibustion robot is a complete system in which programmable machinery positions, moves, or maintains the distance of one or more heat-delivery tools relative to a person, while a controller executes trajectory, dwell, temperature, and safety logic. Designs can include a fixed articulated arm, gantry or linear rail, bedside boom, multi-tool platform, or an integrated unit on a mobile base. The heat source may burn moxa, use an electric treatment head, or combine the two. A powered lift, rotating stand, timer, or remote switch without programmable motion and robot safety controls should not be called a robot solely because of its appearance or marketing.
The object of evaluation is not the robot arm alone. The robot, end tool, heat source and consumable, vision or distance sensors, skin-temperature measurement, smoke capture, bed or chair, positioning accessories, control software, network, and operator form the application system. Repeatability of the bare robot does not establish accurate recognition of a body site, nor does it validate the tool center, thermal exposure, or smoke path after the moxibustion head is installed. System integration and use-setting validation cannot be replaced by a component report.
This entry excludes handheld moxa sticks, passive moxibustion boxes, wearable portable units, fixed tabletop or floor-standing supports, moxibustion beds, and software that only conducts intake or recommends points. Standards for industrial, household service, personal care, and medical robots have different scopes, some with explicit exclusions between categories. Applicable requirements must be determined from intended use, users, environment, mechanical construction, and place of sale; a certificate for one robot category cannot automatically be transferred to a moxibustion application.
Coordinate registration, trajectories, and body motion
A robot executes coordinates and programs; it does not inherently understand a human body. A system commonly relates camera or other sensor coordinates, robot base coordinates, end-tool coordinates, and the individual's body surface. Movement of the bed or chair, deformation of soft tissue, breathing, turning, clothing, lighting, occlusion, or reinstalling the head changes those relationships. Stated mechanical repeatability, visual landmark error, and the final distance error between the heat source and skin are different metrics and cannot substitute for one another.
Confirmation of an acupoint or treatment region also involves anatomical variation and professional judgment. Candidate locations inferred from surface images, a statistical body model, or an earlier record may assist only within a validated range; they should not bypass operator confirmation of identity, site, side, posture, skin condition, and excluded regions. The system should display and record the final target, coordinate source, tool offset, and version, and allow a safe preview before heat is enabled. Tool, camera, bed position, or software changes require recalibration as specified.
Trajectory acceptance cannot be limited to a motionless mannequin and one point. Testing should include workspace boundaries, joints near their limits, intersecting paths of multiple tools, entry of bed rails and accessories, soft-tissue height differences, and modest user motion. Speed, acceleration, dwell time, stopping distance, and replanning behavior all matter. If a body leaves the confirmed region, landmark confidence is insufficient, the view is obstructed, or coordinate sets disagree, the unit should remain in or enter a safe state instead of estimating a position and continuing its approach.
Mechanical approach, collision, and the stopping chain
Mechanical hazards include impact or pressure from the end tool, pinching at joints or rails, shear zones between an arm and bed or chair, cable and hose entanglement, and instability of a stand or mobile base. Limits on speed, force, or power have meaning only for the corresponding structure, payload, end-tool shape, contact region, and measurement method. A rounded cover, software limit, or “collaborative” label alone does not establish safety for close public access. When the tool carries a hot component or ember, even a low-force brief contact can injure skin or ignite fabric.
Risk control needs a layered chain that can include mechanical travel and speed limits, safety-related position or proximity monitoring, safeguarded zones, end limits and brakes, protective stopping, an accessible emergency stop, and controlled reset. An ordinary pause, touchscreen back button, and app stop are not synonyms for an emergency stop. Emergency stopping is also supplementary: it may halt motion without immediately removing residual tool heat, combustion, smoke, or gravity-driven descent. The state of the heat source, fan, brakes, and end tool after an emergency stop must therefore be defined.
Power loss, communication loss, a failed drive, abnormal encoder, brake release, joint deviation, tool detachment, and obstruction of a safety sensor all require fault testing. Reset should not automatically restart a trajectory or heating; the operator first checks the person, tool, consumable, and coordinates before an intentional start. Manual teaching, service, and cleaning modes need restricted authority and motion, with prevention of a remote start from another interface. Hazardous energy should be isolated under a defined procedure before entering the robot's motion range, defeating a safeguard, or clearing a jam.
Thermal output, combustion, and smoke control
Delivered heat depends on heat-source type, head-to-skin distance and angle, trajectory speed, dwell and repetition, path overlap, consumable state, ambient airflow, and heat removal by the body. Rated power, combustion temperature, a head-mounted sensor, non-contact thermometry, and skin temperature are not the same quantity. An average from a moving scan can hide a hot spot where the robot turns, slows, or overlaps paths. Each program needs warm-up, overshoot, uniformity, cumulative exposure, and cooldown verification with the actual tool, load, and posture.
Closed-loop temperature control must identify sensing location, field of view, sampling and response, controlled actuator, and the strategy for sensing failure. Infrared measurement is affected by distance, angle, emissivity, perspiration, clothing, smoke deposits, and occlusion; a single skin point cannot represent a large region or multiple tools. Mechanical distance control and thermal power control can interact, so delay, oscillation, and conflicting limits must be handled. Separate overtemperature protection, maximum dwell, excluded zones, and manual stopping should not all depend on the same camera, software process, or network.
A combustion robot still produces particles, gases, sparks, ash, and residual embers. A tracking hood, local extraction, and filter treat only smoke that is actually captured; fan operation or low visible smoke does not demonstrate zero emissions. Cross-drafts, bent hoses, robot posture, and filter loading change capture. Ignition, automatic consumable exchange, opening, tipping, and extinguishing require coordination with motion controls and continuous operator supervision. At the end, embers must be fully extinguished and cooled as instructed. The system is not for operation around supplemental oxygen, combustible bedding, or a sleeping person.
Workflow, interaction, and data governance
A defensible workflow begins with confirmation of the user's identity, site, and plan, followed by inspection of the equipment, bed or chair, tool, consumable, ventilation, and surrounding clearance. Individual registration is then completed, and the operator confirms target, boundary, distance, and parameters. First operation, a tool change, or recalibration may warrant a path check without ignition or heating. Once started, the operator continuously observes both person and robot and preserves access to the emergency stop and heat-source removal. Coughing, sudden movement, repositioning, or discomfort calls for stopping, not forcing trajectory continuity through automatic tracking.
The interface should clearly distinguish standby, calibration, preheating, motion, dwell, pause, protective stop, emergency stop, and fault, rather than communicating a high-risk state only through color, sound, or similar icons. Privileges for routine operation, plan editing, calibration, servicing, and remote support should be separated, and the system should log who changed coordinates, speed, heat, duration, or limits and when. An alarm needs a stated cause and recovery condition. Masking a sensor, repeatedly resetting, or enlarging a software boundary is not an acceptable shortcut to finish a session.
Vision-based positioning may capture faces, exposed body regions, skin images, and room video, while logs may contain identity, site, treatment plan, temperature, and operator records. Purchasing should establish data minimization, local or cloud processing, access control, retention, export, deletion, and end-of-service handling. Connectivity, remote diagnostics, and software updates must not weaken local stopping, thermal protection, or safeguards. The system should remain safe during network loss, server unavailability, clock error, or an interrupted update, and positioning, motion, alarms, and permissions require revalidation after relevant model or software changes.
Clinical evidence, user limitations, and regulation
A randomized pilot study of 70 people with primary dysmenorrhea compared one moxibustion robot with manual moxibustion. Completion and adverse-event findings offer preliminary information only for that device, point, course, operators, and study conditions. A separate portable-robot investigation reported vision-based positioning error and thermal-feedback control, but it was an investigational system. Engineering accuracy is not clinical effectiveness and does not establish the same performance at other facilities, body sites, software versions, or commercial models. Procurement should separate engineering verification from clinical outcomes.
A systematic review of moxibustion for primary insomnia found high risk of bias, small samples, and limited adverse-event reporting, and the included studies were not trials of autonomous robots. General moxibustion research, manual treatment, a dysmenorrhea trial, or positioning accuracy cannot therefore be extrapolated as evidence that a robot treats insomnia, increases deep sleep, improves circulation, or detoxifies. A study of pre-sleep relaxation would need to predefine the robot version, heat source, consumable, body site, program, and comparator, and record technical failures, adverse events, and sleep outcomes separately.
Broken or infected skin, materially reduced sensation, inability to report discomfort or withdraw promptly, abnormal thermoregulation or circulation, sensitivity to smoke, and conditions that may be affected by heat require product-specific contraindications and qualified professional judgment. Children, pregnancy, frail users, and people needing special care cannot simply receive a standard adult program. A product intended to treat disease requires verification of its medical-device status, registration or authorization, scope, and risk-management requirements in the place of sale. “Robot,” “AI,” or use at a professional facility does not itself create an authorized medical use.
Acceptance, maintenance, and retirement decisions
Site acceptance should use the delivered robot, tool, bed or chair, software, consumable, and smoke system, and review workspace, floor anchoring or base stability, power and protective earth, lighting, ventilation, network, and movement of people. Beyond normal programs, tests should inject body motion, occlusion, wrong coordinates, tool offset, overtemperature, fan stoppage, communication interruption, power loss, and restart, recording the time and distance to stop motion and heat, alarms, and reset behavior. Emergency stop, protective stop, and ordinary pause require separate tests; a factory no-load demonstration cannot represent the worst on-site combination.
Maintenance covers robot and base fasteners, joint play and brakes, encoders and end limits, tool locking, cameras and distance sensors, skin-temperature measurement, heat source and independent overtemperature protection, ash and smoke paths, emergency stops and safety inputs, cables and hoses, and software logs. Accessories touching intact skin and high-touch surfaces need compatible processing between users. Smoke residue, dust, and cleaning chemicals must not obscure sensors or degrade insulation, braking, and flame-resistant materials. Repair, tool replacement, or software change requires recalibration and verification proportionate to the change.
Overall evaluation places truth of body coordinates, the motion stopping chain, thermal dose, combustion and smoke, operator takeover, data governance, clinical evidence, and lifecycle support ahead of smooth motion, arm count, or “automatic acupoint finding” claims. Service should stop and the unit should be isolated when position drift cannot be corrected, a brake or emergency stop fails, a tool loosens, temperature control or capture cannot be verified, faults recur, hygiene cannot be maintained, critical parts or safety updates become unavailable, or the product is recalled. It returns only after qualified repair and re-verification, or is retired under local requirements.