Robots and Intelligent Medical Devices in the Intensive Care Unit: Vision, State of the Art, and Economic Analysis

Robotics is widely seen as a key enabling technology for the society of tomorrow. We examine the role of robotics and intelligent medical devices in Intensive Care Medicine. Demographics predict that more elderly patients will need to be treated with fewer healthcare personnel, calling for innovations in ICU management. Robotics is a key enabling technology for this century and may help smoothing the foreseeable workload/manpower disparity in medicine. While robotics cannot replace the crucial human interactions in healthcare, it’s important to identify those areas where it can contribute in the near and foreseeable future. Studying the application of robotics in the ICU in a manner beneficial for patients and accepted by ICU team is therefore desirable. Financial sustainability is essential for the introduction of new technologies to the ICU. This study assesses the state-of-the-art of robotics in intensive care medicine and identifies opportunities for progress, using an observational approach in a teaching hospital ICU in combination with an in-depth review of the literature and a survey of the market. Tasks potentially amenable to robotics are identified, their acceptability to patients and caregivers are examined and their quantitative contribution to future management of an intensive care unit is assessed.

countries will increase demand for ICM [1]. At the same time, a decrease in the work force (including caregivers) and limited resources available for healthcare call for introduction of innovative technologies in the Intensive Care Unit (ICU). Outcomes in ICM are not only determined by disease severity and therapeutic strategy, but also by economic constraints. The constraints are: availability of human resources: physician availability and education, nursing manpower per bed and support personnel availability (e.g., cleaning tasks to minimize in-hospital transmission of disease). Robotics and automation are key enabling technologies for the 21 st century in many aspects of society and may help to smooth the foreseeable workload/manpower disparity in medicine. Hence, studying the application of robotics in the ICU in a manner beneficial for patients, acceptable by personnel while sustainable financially is therefore desirable. Discussion on healthcare robotics often focuses on [2] surgical robotics for interventions [3], robot prosthetics [4], [5], rehabilitation robotics [6], [7], [8] and social robots for behavioral therapeutics and education [9], [10], [11]. Service robots and robotic technologies embedded in current and future ICU machines are also relevant. Unfortunately, little information is available describing and studying the utility of service robots that care for patients in hospital settings like the ICU, cardiology, neurology and oncology. In addition to patient care, robots might take on tasks of the medical personnel. Smart medical devices and robotics technology already play a role in life-supporting machinery in the ICU. Other medical robotic systems for the ICU are remotely controlled endotracheal intubation robots [12], [13] though their effectiveness have not been proven [13].
A robot is defined as a programable machine that is capable to carry out various tasks autonomously [14]. Formerly, robots were associated with robotic arms doing industrial tasks such as assembly, painting and welding. However, today the term robot is more generic and it also includes autonomous vehicles (mobile robots) and even computer programs that operate autonomously to do virtual tasks.
Artificial Intelligence (AI) is a computer program that is capable of doing tasks that require human intelligence. AI can be based on various approaches such as decision trees, fuzzy logic and machine learning. Machine learning allows building computer algorithms based on training data and inference without explicit pre-programming to do a specific task.
The big challenge in nowadays robotics is coping with unstructured dynamic environments (e.g., the ICU) that cannot be programmed in advance into the robot. The use of AI in a robot (i.e., artificially intelligent robot) is a solution for this challenge. In our opinion, AI will play an important role in future robotic systems. We estimate that artificially intelligent robots will have a major role for carrying out various healthcare tasks in the future ICU. Intelligent medical devices (IMD) are diagnostic and therapeutic medical devices that incorporate AI in their command and control unit. Such devices also called smart devices have a certain degree of automation and they are capable of reacting to variations in external conditions.
The COVID-19 pandemic emphasizes further the importance of robotics and automation in intensive care [15], [16], [17]. Tele-robotics and remote care enable the reduction of the direct contact with the patient and reduce the fatigue of the ICU personnel. While ICU personnel are required to wear protective suits in the vicinity COVID-19 patients, the ability to do the tasks remotely will reduce this inconvenience and the danger of COVID-19 infection. Disinfecting service robots may be particularly beneficial in Covid units because of the higher ICU occupancy [16]. Disinfecting robots are able to increase disinfection intensity while reducing exposure of cleaning personnel. This paper analyses the role of intelligent devices and robotics in the future ICU. First, we analyse the ICU activities in which robotics and automation does or could play a role. Then, we review current research in the listed activities. We also suggest technologies that might be adopted with great effect on ICU needs.

A. Methods
The paper summarizes the state of the art of robotics for the ICU on the background of real-life practice in a large Intensive Care Unit in a Swiss University Hospital (University Hospital of Basel -UHB). It uses a multi-professional viewpoint that incorporates physicians, engineers, nurses, and specialists in hospital economy.
It starts with a concise presentation of current medical practice in the ICU, complemented by an engineering view on technologies suited for ICM and the challenges associated with their use. A survey on available robotic technologies and the scientific literature is presented. We analyse first the field of robotics and then other advanced technologies and medical devices (Intelligent Medical Devices and AI) that help overcome ICU challenges. An economic analysis based on real-life workload and costs arising in medical practice is done to identify where and how robotic technologies could be implemented with the largest benefit.

B. Monitoring and Diagnosis
Because of the immediate vital risk of patients in the ICU, sensing and monitoring technologies are particularly critical. Attention needs to be paid to system robustness, software quality, device protocol for of technical faults and critical patient states and clinical testing. Stringent requirements for device quality, testing and certification may be one reason why robotics and IMD in the ICU is not yet as prevalent as in surgery or patient rehabilitation. Other reasons are the complexity of the ICU diagnosis and the need for independent operation of the devices. Meta-analyses [18], [19] indicate that telemedicine may reduce ICU length of stay and, potentially, mortality. For example, Vespa et al. [20] studied the application of telepresence robots (a mobile robot enabling two way communication between the physician and the patient) in a neurologic ICU. Tele-medicine reduced the latency of the response to brain ischemia and elevated intercranial pressure, increased the occupancy of the unit and saved up to 1.1 M$ per year due to shorter ICU length of stay. Other studies suggest that the acceptance of telemedicine by the ICU staff is high (in most studies the majority of the users reported that telemedicine improves ICU work) [21].
In addition to tele-presence the robotic cameras can be used for patient monitoring too. Advances in image processing enable detecting cues from the patient such as movements and facial expressions. Martinez et al. [22], [23], [24] studied computer vision methods for monitoring patient's sleep in the ICU. The BAM (Bed Aligned Map) system using a set of cameras was studied for estimating patient agitation state and sleep position [22]. The sleeping position was identified reliably, however agitation was difficult to assess. Addison et al. [25] used a regular camera to assess the anesthesia level in a porcine subject and with reasonable accuracy monitored heart rate, respiratory rate and oxygen saturation using image processing of a video stream. The experiment was conducted in supervised conditions and the robustness of the methods needs to be tested in more realistic conditions. Along with monitoring and sensing individual variables, data integration is key for deriving treatment strategies. An ICU physician may encounter up to 200 variables per patient. She/he needs to process this data and make decisions upon it [26]. Consequently, clinical information systems (CIS) are needed to summarize, integrate and store the data over time, examine trends, and put the information in context with other data relating to the patient, with drug and their interactions, and with the environment. Trends and events collected in CIS are not only used for monitoring the patient but to some degree also yield important information on the quality of the acquired medical signals. Detecting patient-critical events while avoiding false positives reliably are key features because frequent false alarms lead to alarm fatigue, hence endangering patients.

C. 'Human in the Loop' Versus Closed Loop Control
Most medical systems use feedback control with a controlling "human in the loop". A medical system measures and displays a value or an alarm, and the nurse or physician interprets the result and, if necessary, adapts a setting or triggers a therapy. Fig. 1 illustrates typical setup of various diagnostic and treatment devices of an ICU. A large amount of patient information collected automatically or manually by the Patient Monitoring System and presented on one, or more graphical user interfaces to the physician and the nurse. Bedside monitors, central monitors, and individual devices collect and process data and forward it to a patient data management system (database). A variety of alarms can originate in any of the devices. Most treatment devices are controlled manually by physician order or predefined protocols, and the nurse registers treatment parameter changes by adding an entry in the database through the user interface.
Human decision is subjected to many reasons for errors, including lack of training, unawareness of critical data, limited decision time, inability to handle large amount of data and limited attention. Clinical information systems will increasingly be connected by a command and control unit to the clinical treatment systems and evolve into closed loop control systems for patient management. By removing routine activities from the workload, such systems promise to give the caregiver more time and focus for those key tasks and decisions where a human can contribute most. Observing that technology is moving towards autonomous control by smart systems, we study the viability and challenges of this vision.
Scientifically, predicting and guaranteeing the behavior of complex nonlinear systems is challenging and poses questions of safety and ethics similar to autonomous cars, airplanes and space vehicles. As medical application of autonomous systems may become widespread, careful analysis of risks is mandated.
Closed loop control is a typical feature of robotic technologies. Closed loop control without controlling human-in-the loop arises in many automation tasks, high precision machinery and mechatronic systems and incorporates a computational unit, sensors and various actuators enabling real-time feedback. Closed-loop control is not only useful at the single device level but the principle is attractive at the overall treatment level, too. In medicine, only few systems connect sensing and treatment in real-time closed loop fashion at the overall treatment level. Examples of such medical systems are ICU ventilators with adaptive weaning protocols [27], image guided surgery [28], and experimentally, robotic needle insertion [29] using ultrasound control.
Programmable syringe pumps are very widely used in the ICU, yet they cannot prevent reliably adverse drug events [30]. Controlling the pump rate by a sensor measuring the biologic target variable may avoid such events. This has motivated the development of closed-loop based anesthesia systems [31]. First systems originating already 60 years ago [32] and significant progress have been made since then. Progress in sedation level measurement, pharmacokinetic modeling of anesthetic drugs, design of controllers and testing of automated drug delivery systems [32], [33], [34], [35]. The challenges for full 'auto-pilot' like closed loop-controlled sedation controller are still considerable. Sedation control is a representative example of the challenges in closed-loop ICU management. First, the depth of the anesthesia is difficult to determine and many clinical feedback signals are used: heartbeat, blood pressure, pupil dilation, lacrimation and sweating and EEG [36]. Second, the human body is a complex system and modeling it for control design is difficult [34]. The problems in modeling are the uncertainty and variability between different patients and the determination of the control variables. Currently the closed loop anesthesia problem is not fully solved [32], [35]. No fully autonomous system known to us has achieved regulatory approval as a commercial product with fully documented effectiveness [37] and safety [38].
Other closed-loop systems for medication are blood glucose level automation using insulin pumps [39], ventilation controlling oxygenation in infants [40] and continuous blood pressure regulation by vasopressor administration [41].
An important aspect of monitoring and treatment is an efficient communication using a protocol that all the medical devices can connect to. This problem has been addressed by IEEE (Institute of Electrical and Electronics Engineers) by publishing IEEE 11073 SDC (service-oriented device connectivity) communication protocol for point-of-care medical devices [42]. The relatively new standard is applied in the ICU [43] and OR [44].

D. Infrastructure Support Activities
The supporting infrastructure of the ICU is also important for successful treatment. Several activities that are done manually by ICU personnel can be automated and aided by robotic system.
One of the most time-consuming activities in the ICU is medication preparation. Several automated drug dispensing units and automated storage units have been developed for medication preparation (See Table V in the following sections). Risør et al. [45] showed that an automated medication system reduces the chance for dosage error by 57%. Automated medication preparation increased slightly the cost of medication preparation however it is still cost effective [46] (economic analysis showed that the additional cost per avoided clinical error was 2.91 Euro).
Service robots in the ICU can be used to bring drugs to patients, deliver invasive samples (blood, urin, etc.) for lab examination and deliver documents to personnel. These courier robots were tested in a clinical scenario in the ICU [47] and in tertiary care unit [48]. The robots delivered drugs from the pharmacy to the patients. In the ICU [47], the robots were able to improve the medication-use process and timelines of medication departure from the pharmacy. In the tertiary care unit [48] there were similar results, though the operating of the robots was difficult (delays because of obstacles) and the staff were not satisfied with them. It was decided not to expand the courier robot test over the two units that were tested. Courier robots have been developed as commercially available products as listed below.
Cleaning and disinfection are very important in the ICU. Several cleaning systems with robotic features are commercially available (see below). The cleaning systems are based on UV radiation and the automated features are safety sensors and estimation of UV dosage needed to disinfect the room. Although the efficiency of the UV lamps has been proven, its use as a cleaning robot has not been clinically tested yet [49].
Various robotic devices have been developed for assistance in moving the patients in bed, aiding nurses. In addition, bed moving robots were developed aiding transportation [50]. To the best of our knowledge these devices (see below) are not yet in regular use and their contribution has not been tested.

E. Economic Viability
A key factor for implementing the suggested technologies, is proving their economic potential. Robotics and automation are getting cheaper with the increase in their use. The most widespread medical robots are surgical robots. Costeffectiveness studies of robot-assisted surgeries compared to laparoscopic surgeries show increased benefits to the patient but also increased costs [51]. It depends how we evaluate the importance of the benefits compared to the costs. However, surgical robots are not necessarily the best example for robotic tasks in the ICU. Robotics tasks in the ICU are closer to robotic logistic systems such as the warehouse robots of amazon [52] (that already proved to be economically beneficial). Personnel time is one of the largest parts of the ICU treatment cost.
Automated medicine preparation systems have been implemented in hospitals and specifically ICUs [53]. Chapuis et al. [54] found out that automated-drug dispensing systems showed a high return on investment: 126,188 Euro saved in year two after implementation and these savings were maintained in the following 4 years. Kheniene et al. [55] also showed that an automated medicine distribution system reduces the medicament expenditures and saves nurse and pharmacy working time.
The largest economic impact of utilization of robotics and automation in the ICU is the shortening of the time devoted by nurses and other personnel for an ICU patient without compromising the quality of treatment. In this study, we will examine the economic potential of the suggested technologies in order to promote cost-efficient high-quality future medical care. Medicine preparation is one of the most time-consuming tasks in the ICU.
Latest statistical analyses [56] of medical decision-making models show that the initial health-status of the patient does not affect the treatment's success, therefore it is worth to invest into moribund patients. This implies that introducing advanced technology for treating patients is economically viable.

II. ROBOTICS IN THE ICU
Robots in the ICU can be used in interventions, nursing, rehabilitation and services. Robots' roles in medicine have been discussed extensively in the last 30 years [57].

A. Robotic Challenges in the ICU
In Table I we tabulate the ICU challenges that can be addressed by robots. In order to illustrate the availability of the devices and technologies we added the Technology Readiness Level (TRL) [58] of the solutions and future trends. We use the definitions of the European commission for the EU Horizon 2020 program [58].

B. Current Robotic Systems
Here we summarize currently available technologies in automation and robotics that are in development or even commercially available.
First, we address robots treating directly the patient. Patient treatment robots are mostly in development. These robots are summarized in Table II. Service robots for logistics are widely used in automated warehouses. The ICU is characterized by the wide use of materials and consumable equipment during the treatment. Medical courier robots have been developed by several companies. Robots that are relevant to service and courier tasks are summarized in Table III Cleaning and sterilization are very important in the ICU. The ICU patients are more sensitive to infection then patients in other units of the hospital. Several cleaning robots  Preparation of medicine is a hard task that requires knowledgeable and experienced personnel. Medication errors may cause complications as we discussed earlier. Automated drug dispensing systems have been developed and applied in the ICU. Table V presents the different automated drug dispensing systems that are available and applicable in the ICU.
The robots and robotic systems in Tables II-V were developed directly for use in hospitals and the medical industry. Robotics systems from other fields can be also applied in the ICU. Automated storage combine with courier robots can be used to deliver rapidly consumables and equipment to the patient's bed and to save time for the nurses. A mobile robot with an arm such as Moxi in Table III, can be used for acquiring samples from the room for early detection of infection causing bacteria.

A. IMD Challenges in the ICU
Similarly, to Table I, in Table VI we discuss the ICU challenges related to sensing, person/machine interface and acuation that are related to introduction of advanced technologies

B. The Future ICU
The ICU patient is usually connected to a large number of devices used for monitoring and treatment. Fig. 1 illustrates all possible medical devices that are in contact with an ICU patient (In reality only a part of the devices is used). Most of them are managed manually. This instrumentation's complexity substantiates the use of automation for simplification of its management.
Part of the monitoring systems are connected to a Clinical Information System (CIS). The patient monitoring system reads out and presents the vital signs of the ICU patient on the main display and the bedside display too. The other part of the CIS is a database. The diagnostic monitoring system collects and records the data from the EKG (heart signal), Ventilator (respiration rate, CO 2 percentage), Sphygmomanometer I (blood pressure) and Pulse Oximetry (SPO 2 percentage). In addition, patient data entered manually by the personnel via the user interface (keyboard, mouse). Current concepts (Internet of Things, IoT) and technologies enable automated computer control of the treatment devices that are connected to the patient. These concepts are utilized by introducing new data transfer protocols [42] that enable application. To improve ICU patient treatment, evolution of ICU equipment may converge towards: 1. All monitoring devices communicate with the clinical information system, a goal that is still far from being achieved in most ICUs.
2. There is still an ample opportunity for more complete monitoring strategies, as real-time neurologic function monitoring, breath analysis, continuous chemical monitoring, continuous flow/composition sensing of urine output and many more. There are many parameters that are determined visually by the medical staff like awake state, stress and sedation levels, delirium states, and such parameters may be amenable to computer vision and may provide quantitative, continuous information, in contrast to today's qualitative, subjective assessment.
3. Sensors that are embedded in fabric, also known as wearable sensors [127], might be suitable to work towards "intelligent" beds, sheets, clothes with early warnings for pressure locations and risk of decubitus, monitoring of sweating and heat distribution and continuous weight monitoring of the patient 4. The IV and arterial lines, that are connected percutaneously for long time periods to ICU patients, are a hazard factor for bacterial and fungi infection. Sensor-equipped IV lines may contribute to fluid flow monitoring and early detection of line-associated infection that is still a significant source of morbidity in ICU patients.
5. Full integration of treatment devices with the CIS eliminates human error and renders the task of entering manually the status information unnecessary. Infusion pumps transferring data on medication rates are a good example for this process.
6. Control of the treatment devices is manual today, but an evolution towards more autonomous systems is currently visible. Examples of such systems are automated ventilation modes, adaptive circulation support devices like ECMO and Impella, and body temperature control devices. This may evolve towards more comprehensive Clinical Treating Systems (CTS) that complement the CIS (See Fig. 2). The syringe pumps are evident candidates for such CTS because of the necessity of adaptive dosage of drugs, e.g., in circulatory instability, although decades of research into closed loop control have not yet found broad application in the clinic.
7. The monitoring and treatment devices that are connected to the CIS and CTS may be incorporated into closed loop control systems as illustrated in Fig. 2, with the CIS providing feedback that controls the CTS. Candidates for such approaches include closed-loop control of the sedation with inputs from electrical brain signals or camera-based inputs [128].  economic aspect of ICU treatment is an important issue that needs to be addressed. In this section we estimate costs that might potentially be saved by introduction of automation and robotics based on real-world data from the ICU of the UHB. To identify opportunities for use of robotics and automation in the ICU and explore its potential economic impact, we analyzed all hospitalizations of patient in the medical ICU of the UHB in the year 2018. Disease categories triggering ICU treatment were: cardio-vascular (53%), neurological (14%), respiratory (14%), metabolic (6%), gastro-intestinal (5%) and other (8%).
The patient stay duration in the ICU is characterized by large variation, suggesting a mix of two major components as illustrated in Figure 3.
To find scenarios where robotics and IMD in the ICU may have the largest impact, we analysed the task distribution of nursing. Nurses are responsible for most of the interaction with the patient and there is a shortage of personnel resulting in work overload. Reducing the nurses' repetitive tasks will improve their work satisfaction and quality.
The main nurse tasks are summarized in In economic terms, automation of monitoring reduces nursing workload and potentially save 918,350 CHF per year. Next, we consider consulting with the physician (Task 2) that corresponds to 4,675 hr or 655,478 CHF. Modern measures of telemedicine saves time by in-situ conferences without physical meetings, nevertheless interaction of nurses and physicians appears important and it is hard to estimate if and how much time such measures might save.
To the best to our knowledge, the literature on the potential time saved by using robotics and automation is very limited. We did the savings estimation by interviewing nursing experts regarding the time needed for each task and comparing them to similar tasks done in other robotics areas (automated storage, service robotics). Note, that this comparison was done in a Swiss hospital therefore the personnel costs is higher compared to other countries. This difference can influence the monetary savings detailed in Table VII. Medication preparation (Task 3) is time consuming. Automatic medication preparation tools are already commercially available (See Table V). Assuming that introduction of infusion preparation and medication sorting systems might save half of the preparation time, such automated processing could save 596,300 CHF per year.  Setting up diagnostic and treatment devices (Task 6) could be improved by automation and courier robots (Table III), but estimating time savings is difficult. Similarly, preparing the patient's place (Task 7) might be improved by automated storage systems with courier robots but will require handson experience for solid estimates. Cleaning robots (Table IV) may have a role and might be particularly valuable in reduced hospital antibiotics-resistant pathogen transmission or highly susceptible patients.
Blood test handling (Task 5) might benefit from miniaturized, automated or continuous blood sampling and transfer. Manual handling of blood tests is time consuming, with an estimated workload reduction for nurses and lab technicians of 66% corresponding to savings of 454,373 CHF.
Non direct handling of urine and defecation (Task 4) can be shortened by courier robots. We estimate that evacuating containers by robots can save 33% of this task (could save 104,424 CHF). The most time demanding task (Task 1) in Table VI is the observation of patients (13,907 hr). The main tool for documentation is the PDMS. Improving the capabilities of patient monitoring by additional sensing systems, smart alarms, AI for patient data evaluation more user-friendly interfaces may enable monitoring to increase nurse productivity without worsening work quality.
Robots aiding patient positioning in bed and moving them out of bed are in development worldwide (See Table II). Such robots can aid in tasks such as: 8. positioning in bed, 9. washing and clothing, 10. patient mobilization and 11. cleaning the patient. In addition, the robots may reduce the physical effort needed by the nurse in patient handling, safeguarding her/his health. Robotics may reduce task workload by 20% (savings of 402,445 CHF). Tasks that require direct contact with the patient's body and complex manipulation, such as: 12. Patient feeding, 14. introducing and retrieving penetrative devices 15. bandaging and 17. participation in intervention, are difficult to replace with robotic technology today. In tasks that require personal and direct communication with the patient such as 18. direct supervision and 19. instructing the patient, robotics and automation cannot replace the nurse.
Handling the belongings of the patient (Task 20) seems not relevant to robotics and automation.
Administration and documentation (Task 21) can be partly shortened by automatic report draft generation by the PDMS and advanced human computer interfaces like speech-to-text. Such tools might reduce time requirement for such tasks by 10%, potentially saving 66,471 CHF.
Overall, robotics and automation have an estimated savings potential of up to 2,570,950 CHF, corresponding to 24% of treatment cost.
Another example for saving time by robotics is implementation of robotic storage system also known as automated warehouse in the ICU. To estimate whether it is effective to implement a robotic storage system in the ICU we analyzed the current needs for bringing items to the patients. From the items we exclude medication and food and take into consideration only medical devices and consumables.
In 2018, 112,317 items were used to treat ICU patients in UHB. 41,403 items were brought from storage and 70,914 items were custom items that were ordinarily placed next to the patient. We assume that the time to bring an item from storage to the patient is 2 minutes. A total of 82,806 minutes per year is spent on bringing items from the warehouse to the patient's bed, in terms of economic value this action costs 245,105 CHF. Based on interviews with experienced ICU personnel we estimate that implementing an automatic storage system could save a minute in bringing an item to the patient. Such a system could save 41,403 minutes (50%) of working time translating to savings of 122,553 CHF in a year.

V. DISCUSSION
Emerging technologies have always had a profound impact on the practice of medicine. Robotics is an emerging field with an impact on most areas of today's society and it is anticipated that it will likewise contribute to the intensive care medicine of the future. While important aspects of current medicine, in particular the human interaction component, will not likely be made redundant by robotics, it is nevertheless strategically important to identify those areas where it can contribute to the quality of medicine and to estimate the economic impact of this technology on intensive care medicine of the future.
Thus, work surveys available robotic technologies for the ICU, identifies areas where future developments may be particularly relevant, and contributes new data facilitating an economic prediction, where robotics development and application may contribute most to the workload and quality of a large contemporary ICU.
New robotic technologies for the care of the sickest need to be feasible, helpful, safe, acceptable to patients and personnel, and affordable/cost-effective. Physicians and nurses in intensive care medicine are typically very open to technological innovation; promising new technologies are quickly adopted, rendering the field attractive to technology innovators.
Humanoid robots are not used in today's ICU yet, but at the same time, robotics, automation technology and AI are rapidly gaining ground. When defining robotics as a combination of mechanical functionality controlled by sensing, computation and closed-loop feedback with a degree of autonomy, some typical contemporary support devices like ICU ventilators, hemofiltration machines and mechanical circulatory support devices already have limited "robotic" features.
Robotic point-of-care diagnostic devices (e.g., for blood gas analysis) may improve patient management by minimizing time-to-diagnosis and reduce personnel workload by substantial device autonomy including self-cleaning, self-calibration and automated quality control.
Transport robots may increasingly help transporting items in an intelligent way from ICU to lab, from storage facilities to ICU, from ICU storage to the patient bed, and might help in patient transportation to diagnostic exams. Although prior experience was not fully successful, [47], [48], since then new sensing systems were utilized [129] and AI aided navigation progressed [130]. The effort of incorporating novel courier robots to the ICU should be renewed. "Low-end" robots like autonomous cleaning machines may be less flashy for a public relations stunt than a high-priced operating robot, however, they have a much larger potential to contribute significantly to hospital hygiene in an era of hospital-transmitted antibiotic resistant bacterial infections.
Care-support robotic technologies for direct patient interaction like in-bed positioning, washing and mobilization, that increase patient autonomy, are still sparse, but are acceptable and awaited by the elderly population in the Basel area [131], [132].
In our workload analysis in UHB ICU, more than 70% of costs are personnel costs and we have identified patient observation, medication preparation, handling of blood and excretions, as well as in-bed positioning and mobilization as the major time/effort requirements for nurses. Support technologies that automate or facilitate some of these tasks are therefore particularly promising economically. Hard numbers that support these aspects are difficult to find in the literature but merit more thorough study.
Economic savings not necessarily means the decrease of personnel. It will enable handling more patients and increasing the capacity of the ICU. This occupational flexibility will enable addressing emergency periods such as the Covid-19 outbreaks in which the demand for ICM was increased dramatically. Having advanced healthcare technologies enables enhanced care without increase of personnel. In addition, in normal ICU operation mode, robotics and IMD will prevent occupational injuries of personnel by reducing their workload.
In many workload-intensive areas, e.g., direct patient care, taking patient samples for laboratory analysis, and medication preparation and delivery in the ICU, robotic technologies are still in their infancy. Rendering predictions about their potential is difficult at the current time.
With the wider usage of robots, patient acceptance will become an issue. Optimal integration of robots to achieve high acceptability requires further study.
Economic aspects play a major role in technology adoption. Generally, high-tech follows the economy of scale, and the global acceptance of mobile phones (with considerable high-tech inside) shows that new technologies may become affordable even in low-income countries when they sufficiently mature. Funds available to healthcare and in particular the ratio between technology cost and personnel cost vary largely in a global and European perspective, and there is therefore a need for cost and workload data from very different socioeconomic contexts and healthcare systems. Recent economic models confirm that increasing severity of disease is associated with increased absolute benefit of medical interventions, even if the relative benefit (percentage of harm reduction) remains the same [56]. Thus, directing financial resources to severe diseases and using advanced technologies like robotics, that come at a price, may prove economically viable as long as there is hope for survival improvement with treatment.
Technological risks and challenges need to be considered. Software and hardware development for safe autonomous systems has proven to be a major engineering challenge, as highlighted in the successes but also failures of autonomous systems in autonomous cars, airplane control, and space exploration. Scientifically, it is not possible to prove in general that a complex nonlinear system will behave "correctly" in all situations that can arise. Digitalization and networking may open the door to intrusions by third parties, in the worst-case leading to loss of privacy or deliberate change of system function. The ethical aspects of autonomously acting machines are still insufficiently studied. Questions of responsibility and liability will need to be solved. On a global level, the north-south divide risks to increase further but also leave a large ground to find adapted high-tech solutions for developing countries that may contribute to existing huge humanitarian problems.

VI. CONCLUSION
While this overview on robotics emphasizes the supportive aspects of robots for human workloads at current and in the near future, the more futuristic questions may be as strategically important and fascinating, but much more difficult to answer: How will AI merge into robotics technologies? In which applications, and at which time will robots in the ICU be capable of not only supporting human activities, but actually matching and surpassing them? Will there be "doctor-less" and "nurse-less" hospitals in the future, somewhat resembling today's robotic car manufacturing sites where the main human activity is installing and programming the robot? Will a science fiction concept of Intensive Care Medicine, performed by an off-the-shelf home appliance, remain science fiction or translate to reality sooner than we expect? While the latter perspective will be shared by few today, past transformations of society by technology, be it book printing, cars and airplanes, personal computers and computer games have shown that we should be ready for major paradigm shifts in Intensive Care Medicine in the not-so-far future.
The challenges due to the Covid-19 pandemic underline the usefulness of automation and robotics in the ICU.
Our conclusions regarding robots and intelligent medical devices in the ICU are as follows: 1. According to the economic analysis the most feasible automation processes are: patient supervision, medicine preparation and test handling. 2. Robotic courier tasks (test handling, automated storage, medicine distribution) are technologically solved and are ready for application in the ICU. 3. For emerging technologies, the most important task for the ICU is the development of integrated CTS (Clinical Treatment Systems) and connecting them to the CIS. 4. Artificial intelligence will play a significant role in aiding ICU personnel in various roles. 5. Direct handling of patients by robots is immature and will require additional R&D to be beneficial for the ICU. 6. Automation of cleaning is feasible and its main benefit in the ICU is infection prevention.

ACKNOWLEDGMENT
Gabor Kosa thanks Anna Imhoff for providing the economic data.