AI and robotics to help older adults: Revisiting projects in search of lessons learned

2.1.1 The RoboCare experience

The RoboCare project [9] has been one of the first research efforts in Europe specifically focused on the integration of AI and robotics technology for domestic assistance services. The result of our research is a prototypical intelligent home (the RoboCare domestic environment [RDE]), in which sensors, robots, and other intelligent agents coordinate to provide support in the daily activities of an older adult [10].

The RoboCare technological ingredients. Since the beginning, the key idea pursued in the project was to create an instance of an End-to-End system for assisting older adults. As shown in Figure 2, the RoboCare prototypical smart home comprizes a single mobile robot and an intelligent stereo-camera. The robotic platform was a Pioneer2 equipped with Robot Development Toolkit functional middleware [11] that integrates the capabilities of a simple path planner and a state of the art Simultaneous Localization And Mapping algorithm [12] driven by SICK Laser Scanner data. A separate subcomponent for robot interaction skills was coupled with the robotic platform. The interaction skills were composed of (a) a “talking head” called Lucia, an off-the-shelf software developed at ISTC,^[1] which was also endowed with speech synthesis functionalities based on the elaboration of specific “content files” in the text format; (b) a speech recognition system called Sonic, a tool developed at the University of Colorado [13]; and (c) a simple InteractionManager developed within the RoboCare project consisting of a rule-based system that fires situation-action rules by activating the specific submodules that are under the manager’s responsibility.

Figure 2

The RoboCare Domestic Environment: AI-based technology reasons on monitored activities and issues warnings and alarms through a robotic platform.

These components are integrated with an AI-based Activity Monitor, the aim of which is to react to unexpected behaviours of the assisted person [10]. The general idea of the Activity Monitor is to “observe” the assisted person’s actions and maintain an updated representation of the person’s and the environment’s state. Based on these observations, the system employs its automated reasoning capabilities of assessing whether the person’s activities fall within predefined behavioural patterns defined by caregivers on the basis of the user’s medical needs that can be considered as desirable for the assisted persons. These patterns are represented in the form of flexible schedules, predefined by a caregiver as an initialization phase of the system, which are reasoned upon by means of state of the art scheduling technology. The system’s inference capabilities allow us to project the person’s activities in time and to synthesize plans to compensate anomalies in the assisted person’s behaviour. Examples of such plans are the speech acts through which the system warns a person of inconsistencies, which can emerge as a consequence of his/her current actions (e.g. suggestions to take the “after lunch medicine”). The main focus of the overall smart environment is to ensure, through daily activity monitoring, the adherence of the assisted person’s routines to healthy behavioural patterns issuing suggestions or warning.

Evaluation in laboratory settings. One of the interesting aspects of this project concerned the video-based evaluation made with potential users of the system. In particular, eight videos were produced that described representative scenarios of possible interaction between older adults and the robot:

1. Environmental safety;

2. Personal safety;

3. Support in finding objects;

4. Support in the management of daily activities;

5. Reminders for medicines;

6. Reminder for medical appointments;

7. Reminders of events;

8. Healthy lifestyle tips.

This article rereads the results of the evaluation in the light of the acceptance models of technology providing an overall and original view of the works described in refs [14,15]. In particular, we consider the UTAUT model [16].

The theory (Figure 3) states that there are four key constructs: (1) performance expectancy, (2) effort expectancy, (3) social influence, and (4) facilitating conditions. The first three are direct determinants of usage intention and behaviour, and the fourth is a direct determinant of use behaviour. Gender, age, experience, and voluntariness of use are posited to moderate the impact of the four key constructs on usage intention and behaviour. Under the lens of this framework, a high perception of usefulness towards the robot emerged from the evaluation. It was especially true for those tasks ensuring personal (detection of medical emergency and warning to a caregiver) and environmental (warning of potential dangerous situations) safety, and for tasks aimed at providing cognitive help while supporting in the daily activities. In line with these findings, the persons expressed more intention to use and higher acceptance in case of preserving both environmental and personal safety. Furthermore, a clear distinction between important and unimportant activities to be performed at home emerged. In the activities which are perceived of greatest relevance (personal and environmental safety), the autonomy of the robot in the management of the home environment and in taking decisions proved to be a very useful resource. In the situation involving an emergency, indeed, the preference for the robotic support is higher so as the perception of utility scores. Conversely, with respect to activities which are not considered to be essential in everyday life, elderly people show a tendency to assign a low score on likelihood of occurrence and even lower scores on usefulness and preference. The system has been judged as easy to use and the emotional impact on the respondents was essentially positive describing the system as relaxing, interesting, and absolutely not uncontrollable, gloomy, or dangerous. It emerged also a relation between perceived usefulness and acceptance and personal aspects. More specifically, it seems that people with a better perceived health conditions are more inclined to accept the system and judge it as more useful. Some concerns have also been expressed. Among them there were the worry about financial costs, the risk of not being able to manage technology due to a possible lack of knowledge. Finally, older adults manifested a sort of apprehension towards autonomous decision making and the risk to lose control and becoming dependent of the system and its services.

Figure 3

The UTAUT – unified theory of acceptance and use of technology.

Lessons learned. The RoboCare experience was very useful for deriving lessons learned both from a technological and human factors point of view. The integration of different technology to provide intelligent services was an innovative and well accepted idea. Nevertheless, this experience also highlighted some weaknesses.

Shortcomings: The technology was preliminary and not mature enough for a real deployment. For this reason, it was impossible to carry out experimentation neither by deploying the system in real houses nor by setting up laboratory experiments. This, of course, prevented users from experiencing the technology in their daily routine and consequently the study results from providing additional important information like the impact on the daily living and the acceptance in the long term. Therefore, a key related challenge can be summarized as follows:

The continuity of use is a key challenge in robotics especially if the technology is intended to help fragile people.

In this respect, a robotic platform should be robust enough to guarantee a continuous use in real-world conditions and should also support users in relevant and important situations. Additionally, the evaluation highlighted also the relevance of Intelligent Environments that cooperate to have a more informed and contextualized knowledge of both users’ and environmental status. The involvement of potential users as key driving factors is also paramount together with a multi-disciplinary approach.

In fact, one of the most original contributions of RoboCare was the multidisciplinary approach:

Multidisciplinarity is crucial to build solutions that are technologically solid but also well accepted by users.

The results related to the distinction between important and unimportant activities suggest, on one hand, to have a proper trade-off between the technology complexity and the relevance of support for the older users; on the other hand, it also suggests the requirement to personalize the support in relation to actual user needs.

2.1.2 The ExCITE project

The basic goal of ExCITE was to help develop a technology that could facilitate social interaction of potentially isolated people (in their home or in a health institution) to increase the level of social participation thus diminishing the sense of loneliness.

The ExCITE technological ingredients. The project exploited a remotely controlled robot, named Giraff (http://www.giraff.org/), able to move within the environment and endowed with a teleconferencing system (Figure 4).

Figure 4

The ExCITE project: testing a telepresence robot in ecological settings and with a long term perspective.

Giraff is a remotely controlled mobile, human-height physical avatar integrated with a videoconferencing system (including a camera, display, speaker, and microphone). It is powered by motors that can propel and turn the device in any direction. The head unit can tilt and pan to simulate eye-to-eye contact using servo motors. The base moves using a differential drive movement system. A remote user can charge the Giraff by driving it onto docking station. The docking station charges the batteries in under 2 h. A full charge is sufficient to allow the Giraff to wander untethered for over 2 h.

The robot is operated by a person (i.e. a relative, a caregiver, or a friend), the secondary user, who wants to contact the senior, namely the primary user in his/her living environment (e.g. home, health care institution, etc.). The secondary user can drive the robot through the primary user’s environment to reach different locations, by using a simple software installed on a computer. The ExCITE project was based on the idea of using a commercial robust robot to face the previously mentioned challenge of continuity of use. In this respect, no additional AI technology was tested within ecological settings. The intuitive idea in ExCITE was in fact to employ simple but robust technology, offering limited but consolidated functionalities, and to investigate possible problems/challenges that may arise from a deployment in real contexts.

Long-term ecological experiments. One of the original features of ExCITE consists of realizing long term experiments involving older adults hosting the robot in their living environment both to communicate with others and to receive assistance services. Figure 4 gives a general idea of the designed method to evaluate features over time. The evaluation entails a period of N months (with 3 ≤ N ≤ 12 ) during which the end user had the robot at home and the clients could visit him/her through it.

Assessment happened at milestones T 0 . Specifically, after an initial assessment (T0 in figure) at the beginning of the experimentation (baseline), the variables of interest are measured at regular intervals ( T 1 − T n ) to observe changes over time. At the last month, the Giraff was removed from the end user apartment and the same variables were assessed again after 2 months from this removal (follow up). The general idea was to use a repeated measures method to see changes over time during the long term usage of the robot. A detailed description of the evaluation protocol is provided in ref. [17].

Table 1 gives an overview of the primary users of the case study participants and the duration of the associated test site. In particular, eight primary users completed the long term assessment and, more specifically, three in Italy, three in Spain, and two in Sweden. Nine secondary users (aged 23–58 years, M = 42 , 9 , SD = 10 , 2 ; two females, seven males) participated in the long term study. Most of them lived in the same city of the primary user which they were connected to, except for one Italian and one Swedish secondary user. Three secondary users (rwo females and one male) worked as healthcare professionals and took on the role of formal caregivers. The others participants are represented mostly by family members – a grandson and four sons – and by a friend of one of the primary users.

We relied on the Almere model [18] that is specifically developed to test the acceptance of assistive social agents by older users. We studied, additionally, Usability, Attitude, Telepresence (Social presence, Spatial Presence, Social Realism, Social Richness), Affective response, and Perceived social support.

Results. Robotic acceptance for all test sites is either maintained or lowered. This is most likely tightly interlinked with expectations and also the diminished novelty of the unit over time and also to technical problems may have influenced the changes over time. Overall, Giraff appeared to be a good means of communication that conveys a nice sense of warmth and intimacy for the older users. The tested people did not feel invaded their privacy by the Giraff presence and this is also confirmed over time. This means that the robot is generally not perceived as an element of intrusion in the old person’s life.

The Giraff is quite appreciated by secondary users who have the role of caregiver (formal or informal). The expectations towards a system that may potentially provide a social assistance having a service or remote monitoring support role are confirmed during the users’ long term experience. The robot, however, was also perceived as too limited and some additional functionalities were desired together with some further level of control. This is in line with the Acceptance Theory Model that claims that the acceptance of technology depends on the perceived usefulness. In this respect, it is important to produce technology that has the right compromise between utility but also simplicity of use and the right level of complexity to allow a natural interaction.

Lessons learned. Many of the aspects that have emerged in RoboCare have been further investigated in ExCITE. The project aimed at testing the concrete and continuous use of the system in real houses by taking into account both technical and social (human–robot [H–R] interaction) perspectives. This has the twofold advantage of allowing an examination of the use of technology in natural environments of use and for sufficiently long periods of time that allow us to observe variations due, for example, to the effect of habit or the growing familiarity acquired with the technology being studied.

This opens up an interesting perspective for developers, who can receive more concrete and contextualized feedback compared to fast tests carried out in a laboratory environment. At the same time, the study of the socio-psychological aspects of technology on people can exploit the advantage of a perspective of continued use over time. The role of AI becomes increasingly preponderant, requiring for the development of such a system a significant number of people with orthogonal skills.

The Ecological long term experiments able to reproduce as much as possible the actual conditions of use of robotic technology are paramount to assess the robustness and usefulness of technology.

Shortcomings. While the ecological long term experiments allowed us to gather valuable information related to the continuity of use, some shortcomings emerged regarding the perceptions of users towards the robot’s capabilities. Primary users had higher expectations about the robot’s abilities and the services offered by the technology. After an initial enthusiasm related to the novelty effect, they expressed the need to equip the robot with more useful services and a more proactive and engageing interaction. From this we can derive a further challenge:

The intelligent and proactive behaviours of robotic solutions are expected from older users. Technology should exhibit intelligence and usefulness in daily life situation to be judged as use-worthy.

2.1.3 The GiraffPlus project

The GiraffPlus project grounds its ideas on a more mature technology able to guarantee continuous operation and long term experimentation. The main aims of the project were as follows:

1. to develop a complete system able to collect older people’s daily behaviour and physiological measures from distributed sensors, to perform context recognition and long term trend analysis. Somehow the strong simplifications of the ExCITE project were mitigated reintegrating the idea of Intelligent Environment (similar to the RDE in RoboCare but with more mature technology);

2. to organize the gathered information so as to provide customized services for both older adults and their caregivers;

3. to assess in the long term the effectiveness of the services and the acceptance of technology in real settings.

GiraffPlus technological ingredients. Figure 5 shows the general idea of the integrated system which is composed of a layer of sensors (Sensor Network) dedicated to the data acquisition from both the assisted person and the environment. These data are then processed by AI algorithms (AI Data Interpretation) to abstract more relevant information for monitoring people’s routine, such as activity recognition or environmental monitoring, to detect possible dangerous situations. Interpreted data are then translated into user services (Personalized Data Visualization and services), bringing information in the right form to each category of users. The basic benefit pursued by the GiraffPlus system is twofold: primary users can access the information on their own health condition, enabling them to better manage their health and lifestyle; they can receive reminders and suggestions as well as visits from friends and caregivers via telepresence. Secondary users are supported by a flexible and efficient monitoring tool that also provides reports and issues alarms in case of dangerous situation for their assisted person.

Figure 5

GiraffPlus integrated system.

Figure 6 gives the general idea of the Data Visualization, Personalization, and Interaction module (DVPIS) that is responsible for the delivery of such services with two different modules: (a) an additional tool (DVPIS@Home) integrated in the robot at home which delivers additional information to the assisted person in the home environment and (b) a dedicated subsystem called DVPIS@Office for the broad class of secondary users. Both these modules are composed of a back-end part, devoted to organize the content of the information to be shown to the users, and a front-end part, responsible for presenting the information and services to the different categories of users.

Figure 6

Personalized services for primary and secondary users.

Services for primary users. Services for the older adults have been conceived according to the ethical principle that the primary user as “data producer” should maintain a level of understanding and self-awareness on their health and should also have a means to keep in contact with the external world. The main services provided to primary users are therefore:

1. Avatar: This functionality preserves the “traditional telepresence” service that the Giraff robot provides so that secondary users can visit the older user’s apartment through the telepresence robot.

2. Messages: An additional environment has been added to allow the primary user to receive messages from secondary users or reminders and suggestions. Messages and reminders are provided in both textual and spoken form.

3. Shared data: This environment allows us to show personal data to the primary user (e.g. physiological measures) and endows the system with a shared space between the primary user and the secondary users that could foster a discussion on the health status and habits of the old person. The general aim is to improve older adults’ awareness and also to encourage responsible behaviours for increasing their well-being.

It is worth highlighting the use of automated planning techniques for the administration of personalized and adaptive reminders [19]. Automated planning, in particular, is used to plan “virtuous paths,” to be followed by the primary users, to achieve certain objectives, determined by the secondary users, related to the primary users’ personal well-being.

Personalization, in particular, takes place taking into account both the current state of the user, inferred through context analysis techniques, and, at the same time, the determined objectives. The result is a plan, characterized by a set of personalized stimuli that pursue one or more objectives. Finally, the execution of this plan takes the form of sending these stimuli by messages about food suggestions and healthy lifestyle hints, like specific seasonal vegetables and fruits or, in summer, advises about drinking water.

Services for secondary users. One of the most interesting aspects of services for secondary users is the possibility for Health Professionals to follow a list of homes (and consequently primary users) that have the GiraffPlus system installed in their apartment. Specifically, the list of followed user contains a brief and immediate information on the status of the assisted person in terms of three main indicators: Alarms, Physiological, and Social aspects. For each of this dimension an immediate feedback is given with a judgment on the level of each indicator: g r e e n = g o o d ; y e l l o w = w a r n i n g ; r e d = r i s k . In this way, a secondary user can easily judge if he/she needs to urgently intervene on some specific situations and in general he/she can modulate and prioritize the visit to the different patients, thanks to an immediate feedback without the need to entering into the details of each home. For each of these people, the secondary users can observe the real time view of the house on a map depicting sensors installed in the house; the physiological data specific for the patient; can also ask daily/weekly or monthly report for the main activities to observe possible deviations from routines. Another relevant service for secondary users is the implementation of a new panel dedicated to foster the discussion among the network of persons related to a primary user. Specifically, an environment has been developed where the different actors involved in the care of a primary user can exchange information and opinions so as to maximize the overall care for the old person at home.

Long-term experiments. Similarly to the ExCITE project, also this system has been tested in the long term. Specifically, 15 test sites were considered for 1 year of investigation focusing on the following dimensions: Acceptance, Adoption and Domestication, PIADS, and Telepresence. Overall, the perceived acceptance was maintained over time and the Adoption and Domestication dimension confirmed that users appreciated the system functionality and felt it could become a useful tool part of their daily routines. In support to this it is also worth saying that at the end of the experimentation in many cases we decided to leave the system in the home of the experimenters, having also the need to define an “exit strategy” for the end of the experiment. This highlighted the need to have not only the technology but also the equipment of health professionals and care givers as enabling factor for the success of this kind of technology.

Primary user results. From the primary users’ perspective, older adults felt more protected having a monitoring system at home; they felt “more autonomous” and they felt they weighed less on their families. Older adults appreciated the ability to communicate via the robot yet they expected more manipulative capabilities and at the same time asked for additional features (e.g. voice commands). The issues related to privacy did not seem to be a problem even though in some cases the configuration of the sensor network was limited by privacy concerns. Some interest about costs also emerged and older adults expressed concern on who should pay for such a service.

Secondary user results. Family members were very interested in alarms and possible dangerous situations and they also appreciated the periodic report on the older adult’s routine. Doctors appreciated the possibility of routinely monitoring and correlating physiological data with lifestyle habits. Finally, doctors and formal caregivers saw GiraffPlus as an aid to their work (e.g. they appreciated the possibility of monitoring multiple people at a distance, optimising real visits, and better manageing false alarms).

Lessons learned. With the introduction of sensors experienced in GiraffPlus, the need to interpret the data coming from them has immediately emerged. Even if only limited to the possibility of the robot to autonomously return to its charging base or to the ability to send personalized and contextualized messages to the users, more and more autonomy is required from the robotic platform. Also, the capability of providing end users with proactive assistance, “anticipating” their specific needs and habits emerged as a clear requirement in order to perceive the robot actually as an acting and “thinking” entity of the environment. In this regard, robot autonomy should deal with unpredictable and heterogeneous behaviours of end users to support “natural” and reliable interactions. Namely, robotic solutions should be aware of the uncertainty concerning H–R interactions and carry out (continuous) assistance in a robust way.

The Robustness and reliability of robotic solutions are crucial in daily-living scenarios. Technology should exhibits the level of autonomy needed to interact with users and environments in a smooth and continuous way.

The Human-awareness is crucial to realize robotic effective solutions that tightly interact with end-users. The autonomy level of a robot should take into account the possible dynamics of human users in order to guarantee robust and continuous assistance.

Shortcomings. A delicate aspect that had to be faced in the ecological and long term experimentation was related to the conclusion of the experiments and the consequent interruption of the supportive service. In fact, technology alone was not sufficient to guarantee the continuity of a service that inherently requires organizational support behind it to be functional and effective. This has highlighted a new challenge for system design like GiraffPlus which are inherently socio-technical systems, which means that they act as enabling technology for new socio-health assistance models but they cannot disregard the human component.

Technology alone is not enough to provide better assistance but it is an enabling factor to optimize and renew the care and assistance systems which are socio-technical systems.

2.1.4 SI-Robotics

As already briefly mentioned for the GiraffPlus project another important aspect of the robotic support systems concerns the ability to serve different situations and scenarios by providing modular and customizable solutions.

This is what we are pursuing within the SI-Robotics (SocIal ROBOTics for active and healthy ageing) project, whose aim is to design and develop novel solutions for socially assistive robots to support seniors in different contexts.

The objective is to propose novel AI-based robotic solutions realizing a variety of assistance services in different scenarios ranging from daily-home living (e.g. for continuous daily assistance) to hospitals (e.g. for health monitoring or rehabilitation support). The daily self-management of one’s own health, declined in activities such as following a correct diet, practicing constant physical and cognitive exercises and taking drugs adequately, often represents an important challenge for the older adults population, characterized by fragility, cognitive decline, or poor health and technological literacy. Personal robotic assistants, able to promote healthy lifestyles, characterized by an empathetic communication and reliability over time, can help solve this problem by adopting strategies that also aim to motivate the assisted persons.

The SI-Robotics technological ingredients. Central to SI-Robotics are personalization and adaptation which translate into the capability of contextualizing assistive services to the needs and “features” of different social and environmental contexts as well as health needs of assisted persons. The project aims at addressing these issues from both hardware and software perspectives. On one hand, it aims at realizing an innovative modular robotic platform that can be “easily” configured and adapted to different scenarios. On the other hand, the project aims at developing and integrating AI technologies to realize the cognitive capabilities a robot should be endowed with to autonomously recognize features and needs of different scenarios and consequently decide and synthesize effective “tailored” assistive services. The project grounds on an integration of different AI technologies, i.e. machine learning (ML) for long term and continuous adaptation, automated planning (AP) for flexible decision making, knowledge representation & reasoning (KR&R) for context awareness with the aim to support behaviour flexibility.

Current results. Research efforts have focused on the integration of ML, KR&R, and AP to support physical/cognitive rehabilitation scenarios [20,21]. Taking inspiration from dual-process theory [22], we have focused on the design of a novel cognitive architecture supporting the synthesis of personalized stimulation plans, administrated through adaptive H–R dialogues. A feasibility study shows a demonstration of the pursued approach to model different needs and preferences of persons based on their health needs and provides examples of personalized stimulation and H–R interactions modulated according to the model.

More specifically, an ontological representation of the International Classification of Health, Disability, and Health (ICF – https://www.who.int/classifications/icf/en/) has been defined to model health needs of assisted persons and H–R interaction preferences. Knowledge reasoning mechanisms process user knowledge by inferring cognitive (or physical) impairments. According to the refined knowledge a set of tailored stimulation actions are identified and given as input to a planning engine, which is in charge of synthesizing a personalized stimulation plan.

Namely, a planner receives input knowledge characterizing the types of stimulation action that fit the needs of a patient together with a number of interaction parameters characterizing the way such actions should be executed. Personalized stimulation plans are then given to an ML reactive layer that encapsulates Natural Language Understanding features to execute stimulation actions through dialogue-based interactions.

Identified challenges. The SI-Robotics project pursues a more general perspective where a robotic system should support different types of end users in different types of scenarios enabling a modular and personalized support. The pursued approach is to integrate a number of AI technologies to endow the system with the cognitive capabilities needed to autonomously recognize the needs of assisted persons, the objectives of a specific scenario and adapt its behaviour accordingly. One of the main identified challenges is the synthesis of assistive systems endowed with the intelligence needed to contextualize their assistance according to the scenario and the users it interacts with. On one hand, personalization and adaptation concern the types of service users need in a particular context and health-related situation (i.e. what kind of assistance the system should provide). On other hand, these qualities concern the way such services should be carried out to “maximize” the efficacy of the assistance and user acceptance (i.e. how a robot should interact with users according to their features, interaction capabilities and preferences).

Personalization and adaptation of robot behaviours are paramount. General skills and assistive capabilities of robotic solutions should be tailored to the heterogeneous needs and interaction features of assisted end-users.

Another important challenge is also related to the safe deployment in real settings that may present different physical constraints:

The Safety of robot technologies should be considered when deploying them in real-world scenarios. Autonomous behaviours of a robot should be constrained according to the risks concerning the safety of human users.

Indeed, safety is particularly relevant in critical environments such as hospitals or the homes of fragile people.

2.2.1 Robotic solutions for AAL

A large number of research initiatives are dedicated to develop and propose robotic solutions for AAL such as, e.g. SAR [25] or Robotic Socio-Ecological Systems [5]. The projects described in Section 2.1 contributed to the state of the art but, clearly, there are also other work/projects focused on addressing different aspects. Along the same line, several other research initiatives and projects investigated the integration of robotics and ambient intelligence solutions for supporting older adults living alone in different scenarios. There was a strong effort in this direction and here some representatives are presented just to mention few of them.

Robot-Era project [26] aimed to increase the acceptability improving the quality of robotic services proposing a cognitive-inspired robot learning architecture and using different platforms for indoor and outdoor applications. Its scope was to integrate robotics and ambient intelligence technologies, AI and cognitive-inspired robot learning architectures, elderly user-needs, methodology of design for acceptability, and legal/insurance regulations and standards fundamental for the real deployment. All these skills have been applied to generate robotic services for “ageing well” and to propose a set of appropriate solutions to overcome actual barriers to the exploitation of robotic services. The Robot-Era platforms were tested in realistic environments to evaluate their actual impact on acceptability with end users considering different application scenarios, i.e. domestic, condominium, health facilities, and public spaces.

ENRICH-ME [27] aimed to enhance the day-to-day experience of elderly people at home with technologies that enable health monitoring, complementary care, and social support. ENRICH-ME proposed a system composed of a companion robot deployed in a sensorized environment and a Networked Care Platform providing a set of remote monitoring services for healthcare professionals based on smart environments, human perception and advanced autonomous navigation technologies, and cognitive robot control. The system was evaluated in laboratory settings and, with a limited temporal deployment, also in some real environments.

Similarly, CompanionAble [28] addressed the issues of social inclusion and domestic care of older adults with mild cognitive impairments. The project focused on combining the strengths of a mobile robotic companion, called Hector, with the advantages of a stationary smart home. Hector was enriched with remote monitoring services, personalized dialogue/interaction displaying emotional intelligence to avoid feelings of loneliness, provide friendly reminders, store/bring important objects such as keys, wallet, and offer cognitive stimulation/games, as well seamless video connections to family and friends. The CompanionAble system was installed in a number of demonstration homes testing its functionalities.

Mobiserv project [29,30] developed an integrated and intelligent home environment for the provision of health, nutrition, and well-being services to older adults. The goal has been to develop and use up-to-date technology like a companion robot, smart home, and smart clothes in an intelligent and easy to use way to support independent living of older persons. Based on state of the art AI and robotics technologies, the Mobiserv robot companion was designed to offer cognitive support to users, offering reminders and suggestions to help them lead healthy and socially active lives. During some trials, users experienced the system for several hours with a researcher present to observe and initiate a dialogue. The Mobiserv system was then evaluated and validated in a realistic setting with target users spending a week in a smart home.

In GrowMeUp [31], the focus was on allowing elderly people to live for longer in their own environment without losing contact with their social circles, staying active either via teleconference or other social facilities provided within the system. GrowMeUp developed an affordable robotic platform capable of learning older person needs and habits over time to enhance existing or build new services for compensating older person capability degradation and gradually adapting its interaction over time. The system provided several functionalities for end users based on Behaviour and Emotional Understanding, an Intelligent Dialoguing and Personalized care. The robotic platform was validated through the execution of nine use cases at participant end users and both older persons and caregivers found the robot useful and motivating towards a more independent and active life.

RAMCIP [32] aimed to develop a service robot, capable to assisting older persons in a wide range of daily activities, being at the same time an active promoter of the user’s physical and mental health by deciding when and how of assisting the user. The RAMCIP robot comprized three major innovative aspects: cognitive functions based on advanced user and home environment modelling and monitoring, allowing the robot to decide when and how to assist the user; novel adaptive multimodal human-robot communication interfaces, with emphasis on empathetic communication and augmented reality displays; advanced, dextrous and safe robotic manipulation capabilities applied in service robots for assisted living environments.

The aforementioned projects share most of the lessons learned in our path and, thus, reinforce our findings setting them up as general points for developing assistive robotics solutions for elderly people.

2.2.2 Additional lessons learned

Going beyond these results, and in line with the classification proposed by ref. [23], some projects focused on physical impairments, with major attention to mobility issues. At the same time, there are examples of developmental processes whit incremental user involvement through years as, for example, the case of the DALI project [33] and the subsequent ACANTO project [34]. The first one aimed at developing a robotic cognitive walker (c-Walker) that can be taken to, or picked up at, the place to be visited, gently guiding the person around the building safely. The specificity of this device lies on being able to take corrective actions when the user comes across the type of busy area, obstacle, or incident they want to avoid. The second one, funded within the Horizon 2020 programme, represents the evolution of these efforts where the scientists put their efforts in refining the technology deeply relying on a user-centred design and designed longitudinal study to assess the clinical effectiveness of the developed device (the FriWalk) both on physical and psychological well-being [35,36]. These two projects developed a low cost robotic platform capable of analysing end users’ social network activities to understand the social context, implement proper interactions with humans, and endow the system with a recommendation system for user activities.

Robots supporting older adults in maintaining their mobility and helping them to prevent falls are crucial for prolonging independence and preserve autonomy.

A further, and different, aspect addressed in some work concerns emotions. Indeed, emotions are included among the crucial interaction components for a socially interactive robot [37]. Therefore, social robots can be used to encourage emotional expression in situations where it may be challenging. So, for instance, robots are used to encourage children with autism to open up emotionally [38] and empathy is considered as relevant in providing assistance to elderly people [39,40]. With older adults, emotion constitutes a highly relevant human factor to take into account to improve user engagement while interacting with assistive robots. On one hand, psychologists showed that empathy plays a key role for therapeutic improvement and their assumption is that empathy mediates pro-social behaviour (see, e.g. ref. [41]). Patients who have received empathy from their therapist recovered faster and the same seems to hold with assistive robots. Robots can be designed to show empathy to improve user satisfaction and motivation to get better as well as enhance adherence to therapy programmes in the context of patient–therapist interaction [40]. The work in ref. [42] presents a prototype for an emoting robot that can detect emotions in one modality, specifically in the content of speech, and then express them in another modality, specifically through gestures. The robot is able to detect and express emotions through an emoting system. Results from a human validation study show people perceiving the robot gestures as expressing the emotions in the speech content. Additionally, people’s perceptions of the accuracy of emotion expression are significantly effective. In ref. [43], an effective reasoning system has been implemented in the NAO robot for simulating empathic behaviours in the context of AAL. In particular, the robot recognizes the emotion of the user by analysing communicative signals extracted from speech and facial expressions. The recognized emotion allows triggering the robot’s affective state and, consequently, the most appropriate empathic behaviour. The robot’s empathic behaviours have been evaluated both by experts in communication and through a user study aimed at assessing the perception and interpretation of empathy by old users.

The MARIO project [44] aims to help people with dementia by enabling them to stay socially active using touch, verbal, and visual H–R interaction tools. A unique capability of this effort is the ability to engage users with dementia in reminiscing about their past as well as people and places with emotional significance. Such reminiscing approach is well suited with mild cognitive impaired people to overcome communicative barriers and promote social interactions. The key objectives of MARIO were to address loneliness, isolation, and dementia in older persons through multifaceted interventions delivered by service robots, conduct near project length interaction with end users, and assist caregivers and physicians in conducting comprehensive geriatric assessments. Machine learning techniques and semantic analysis supported language, space, and mood recognition making MARIO more personable, useful, and accepted by end users. An assessment of deployment of MARIO robots was performed to provide evidence related to the benefits/impacts of the use of service (companion) robots involving a number of elderly people with dementia and their caregivers within three different environments, i.e. nursing home, hospital, and community and showing a good impact in terms of acceptance, engagement.

On this aspect, it is worth also mentioning efforts made to foster elder motivation in using technology, which has been noted to be a pretty hard challenge in the developing design. It can be the case of the HOBBIT project [45], more oriented to foster the interaction with the robot, whose core zoomed in on the interaction between robot and owner/user through a more user-centred concept called Mutual Care. This project allows and entices people to “take care” of the robot like a partner, so they can develop real feelings and affections towards it. The underlying idea is that, for people, it is easier to accept assistance from a robot when themselves can also assist the machine. Moreover, HOBBIT underwent an iterative assessment process from laboratory setting to ecological environment. The preliminary evaluation sessions in laboratory setting focused on the assessment on perceived usability, acceptance, and affordability of the robot and demonstrated a positive reception of the robot from its target user group [46]. Nevertheless, findings from a long term study show that the robot was rather seen as a toy instead of being supportive for independent living, despite utility met the user’s expectations [47]. To summarize, the above work considers emotions, empathy, feelings, and affections as relevant features that, as in human-to-human interactions, can play a key role in healthcare activities for older adults.

Designing robots capable of detecting and expressing emotions and showing empathy allows to deploy engageing robots and support therapy adherence.

Finally, some work is in progress to develop robots capable of adapting their behaviour according to different cultures. According to the cultural competence concept, i.e. the ability to respond effectively to people from different cultures and backgrounds, solutions can be developed to assist healthcare professionals in the delivery of services that meet cultural and communication needs of patients [48]. Under this perspective, culturally competent robots are to autonomously reconfigure their way of interacting with a user in a way that is appropriate to the culture and preferences of the person they are assisting. Though relevant, this concept has been rarely implemented and evaluated within robotics despite its importance for enhancing patient-centred care. Such abstract concept is implemented, e.g. in ref. [49]. This research project (funded under an EU–Japan cooperation programme for developing advanced robot-based solutions for extending active and healthy ageing and evaluating such solutions under different cultural perspectives) aims at designing, developing, and evaluating culturally competent robots that can assist older people according to the culture of the individual they are supporting. In CARESSES, socially assistive robots are configured to assist older adults residing in long term care homes in a culturally competent manner [50]. Robots understand which culture the user primarily identifies with and leverages a relevant cultural knowledge database to adapt its interaction behaviour. The robot uses this database (a hierarchically structured ontology) as a basis of its verbal and non-verbal interactions but then adapts its understanding of the user’s individual preferences and values as it receives feedback from the user during an interaction. Users may be more likely to accept and value interactions with culturally competent robots. This is important given how critical user acceptance is for the successful implementation of any public health intervention.

Culturally competent robots can configure their way of interacting with users according to their culture and preferences.

As summarized in Table 2, the projects presented above represent a wide amount of work considering many different AI and robotics technologies and solutions, from knowledge representation to advanced autonomous navigation, to address several challenging issues related to fostering the diffusion of socially assistive robots. Most of the technologies often present some limitations. This is mainly due to the fact that such solutions have been often deployed as robotic prototype tested and evaluated in controlled environments for a limited period of time. Apart some very specific cases, the main challenge remains in developing reliable solutions capable of supporting long term use in real scenarios.