Technical Field

The present invention relates to an artificial intelligence system for smart home. Background Art

The need is well known and increasingly felt to create artificial intelligence systems for so-called“smart homes” which are efficient, complete and, at the same time, easy to use.

Nevertheless, currently known systems have a number of drawbacks.

In particular, the systems on the market generally send information collected in the home, such as images, videos and other data collected by sensors, to a remote computer.

As a result, this information passes through the public network, posing a real risk to the user’s privacy.

This problem is particularly relevant in the case of the processing of multimedia information, in particular videos from one or more cameras installed inside the home.

Moreover, systems of known type are generally capable of managing a single type of data (e.g. noise level, presence, movement, temperature).

This does not make it possible to assess the interdependence existing between different types of data, interdependence that can potentially make it possible to obtain semantically more relevant information.

To overcome this drawback, software platforms exist which make it possible to use several data sources. These known platforms do however require complex user- side programming and management.

In addition, the systems of known type generally require masonry works to be carried out to install multiple devices inside the home. Such jobs are sometimes difficult (think, for example, of jobs done inside historical buildings) or extremely expensive.

Moreover, the systems of known type are generally based on an individual gateway architecture. Therefore, in case of gateway failure, the correct functioning of the entire system cannot be maintained. Known systems generally require an Internet connection, in the light of the fact that all the proposed functions can only be performed in the presence of information or algorithms located on a cloud platform. Nevertheless, in the home environment the user does not expect not to be able to use certain functions if there is no Internet connection, especially when these functions are not perceived as strictly related to the Internet connection (e.g. heating control or roller shutter automation).

Description of the Invention

The main aim of the present invention is to provide an artificial intelligence system for smart homes which allows the effective management and analysis of different types of collected data.

Another object of the present invention is to provide an artificial intelligence system for smart homes which allows protecting the privacy of users.

Another object of the present invention is to provide an artificial intelligence system for smart home which can be easily installed inside a home.

Another object of the present invention is to provide an artificial intelligence system for smart home which is fault-tolerant.

Another object of the present invention is to provide an artificial intelligence system for smart home that is able to function correctly and effectively even in the absence of an Internet connection.

The objects set out above are achieved by the present artificial intelligence system for smart home according to claim 1.

Brief Description of the Drawings

Other characteristics and advantages of the present invention will be more evident from the description of a preferred, but not exclusive, embodiment of an artificial intelligence system for smart home, illustrated as an indication, but not limited to, in the attached tables of drawings in which:

Figure 1 is a general diagram of the system according to the invention installed inside a home;

Figure 2 is a general diagram of a single node of the system according to the invention; Figure 3 is a functional diagram illustrating the operation of an automatic reasoner of the system according to the invention.

Embodiments of the Invention

With particular reference to these figures, reference numeral 1 globally indicates an artificial intelligence system for smart home.

In particular, the system 1 comprises a plurality of nodes 2 for the detection and processing of data, which nodes can be located within different areas of a home and operationally connected to each other by means of a wireless network 3. Advantageously, each of the nodes 2 comprises:

- at least one sensor device 4 for the collection of data inside a respective area of the home;

- at least one storage unit 5 of the collected data;

- at least one processing unit 6 with high computational capacity configured for the processing of the collected data;

- at least one wireless communication unit 7 configured for the communication with at least one of the other nodes 2.

Therefore, each individual node 2 is able to collect different types of data by means of one or more sensor devices 4 and is also able to process such data collected by means of a dedicated processing unit 6.

In addition, the system 1 comprises distribution and synchronization means 8 configured to distribute and synchronize the data collected by each of the nodes 2 on all the nodes.

In particular, these distribution and synchronization means 8 are implemented by means of a middleware for the communication between the different nodes 2.

In actual facts, therefore, the data related to each relevant area or home environment are collected on the storage units 5 of each individual node 2, are processed locally by the processing units 6 of each individual node 2 and are distributed to each node by means of the distribution and synchronization means 8.

Therefore, advantageously, each of the nodes 2 has all the knowledge available within the whole system 1, i.e. all the data collected by all the nodes 2. The data distribution middleware 8 is developed so as to adapt to the critical conditions of the wireless network 3.

In particular, the system 1 implements QoS (Quality of Service) and data compression policies to optimize the use of available bandwidth.

Each application running in the system 1 subscribes to the transmission and reception on a subset of messages of predefined types (topics) defined in the system itself and characterized by respective unique identifiers (UUID).

Each application has a local cache that keeps the last copy of the data for which it has subscribed (in accordance with the QoSs of each message, indicated below). Advantageously, the distribution and synchronization means 8 comprise at least one persistence service running on each node 2 of the system 1.

In particular, the persistence service is configured to perform at least one of the following tasks:

- at startup it reads from the storage unit the collected data (state of the home) and publishes them by populating/updating a cache of application level;

- it listens to any messages transmitted on the wireless network and, according to a persistence policy associated with the specific type of message, performs the persistence thereof on the storage unit 5;

- it transmits at least one correction message if the transmission of an old message is detected (due e.g. to the startup of another persistence service on another node 2).

In particular, for each type of message (topic) is defined a set of QoS (Quality of Service) policies regarding at least the following policies:

- reliability: indicates whether it is a piece of data that has to be compulsory delivered to all recipients or whether the message may be lost;

- persistence: specifies the type of persistence of the piece of data with respect to switches-off and restarts of the system and of the services;

- domain: specifies whether the piece of data is also propagated to the cloud;

- access: specifies the authorization policy for the piece of data.

For example, with reference to the reliability policy, on a periodical datum, it may make sense to wait for the next transmission period and not overload the network with a retransmission.

According to a preferred embodiment, the persistence policies comprise the following policies:

- volatile: the datum must not be stored (involves only the reception and sending of requests but not their storage, e.g. signals); such datum is only available in the management cycle of the reception event;

- cache_only: datum stored only in the RAM (volatile) memory of applications and persistence services;

- disk_only: datum stored only on disk (non-volatile), the datum is not maintained in RAM;

- cache_and_disk: the datum is stored on both RAM (volatile) and disk (non-volatile);

- volatile_app_cache: datum stored only on the application RAM and not on persistence services RAM; the applications have the datum in their local cache (can take readings); the persistence services do not store the datum;

- gateway _loopback: the message is not transmitted over the network but is used to exchange messages within the individual application process, to make the process loosely coupled at the architectural level.

According to a possible embodiment, the domain policies comprise the following policies:

- local_and_cloud: the message is sent both locally and in the cloud;

- local_only: the message remains confined to the local network of the home. Usefully, all the messages which contain sensitive information are of the local_only type. Local_and_cloud type messages are only those related to commands (and their feedback) sent through the app.

According to a possible embodiment, the access policies comprise the following policies:

- public: the message can be displayed by both mobile applications (users) and the nodes 2;

- private: the message can be displayed and modified only by the nodes 2. Advantageously, the system 1 comprises a plurality of micro services, made up of their respective software running components, classifiable into two separate categories:

- node services 9 running on each node 2 of the system 1 and configured to collect data strictly related to the area of relevance of the specific node inside the home and by means of at least one sensor device 4;

- cluster services 10 (or“system” services) running on at least one node 2. The cluster services 10 are configured to carry out tasks relating to all nodes (having a“global” feature) and usually carry out tasks which can hardly be performed in a completely decentralized manner.

In particular, the clustering manager is implemented by means of a control algorythm being configured to carry out at least the following steps:

- monitoring the cluster services 10;

- in the event of an error in the cluster service 10 or failure of the relevant node 2, rescheduling the cluster service 10 at a different node 2.

Therefore, each cluster service 10 is constantly monitored and maintained in operation by the local control algorithm which, in case of an error or failure of the same, is able, in a short time, to reschedule them elsewhere.

Advantageously, according to a preferred embodiment, the cluster services 10 are running on a plurality of different nodes 2.

In fact, the cluster services 10 must be running on at least one of the nodes 2 but, for fault-tolerance reasons, it is not advisable to have them all running on the same node 2.

Advantageously, the system 1 comprises an appropriate clustering manager configured to manage the distribution of the cluster services 10 on multiple nodes

2.

In practice, therefore, the clustering manager ensures that the correct set of cluster services 10 is running throughout the system 1.

This particular configuration of micro-services is schematically shown in Figure 1, where on some of the nodes 2 the cluster services are running, while on all nodes 2 the node services 9 are running. Therefore, the scheduling of the cluster services 10 (i.e. the choice, for each cluster service, of the node that will run it) is carried out in a collaborative and decentralized way by the clustering manager.

Conveniently, the clustering manager is running as node service 9 (i.e. on all nodes).

According to a preferred embodiment of the system 1, the node services 9 comprise at least one of the following:

- an image acquisition and image processing service;

- an acquisition, pre-processing and processing service of an audio stream (e.g. for running voice commands);

- an acquisition and pre-processing service of environmental data from at least one sensor of environmental parameters (temperature, CO2, CO, VOC or the like);

- a network diagnostic service;

- a software update service.

Advantageously, the cluster services 10 comprise at least one automatic reasoner. In addition, the cluster services 10 may comprise at least one of the following:

- a weather data reading service;

- an implementation management service;

- connection services to third-party multimedia systems.

The automatic reasoner is configured for the processing of the collected data and for the implementation of predefined activities (predefined rules and/or smart behaviors).

A general functional diagram of the automatic reasoner is shown in Figure 3.

The automatic reasoner is configured to receive at input all the collected data D1 from the input sources 11 , whether they are physical sensor devices 4 or processes that generate data of interest for the correct management of the home.

Advantageously, the automatic reasoner comprises a component of data fusion 12 configured for:

- receiving at input the collected data D 1 ;

- processing the collected data to generate a set of semantically relevant information D2.

The automatic reasoner is also configured to receive at input the user commands C coming from at least one command unit 13.

For example, the command unit 13 may consist of voice command software, mobile app command or physical input.

Advantageously, the automatic reasoner comprises a reasoning unit 14, made up of a state machine 15 and of a reasoning engine 16.

The state machine 15 is configured to receive at input the semantically relevant information D2 together with the user commands C.

In particular, the state machine 15 is configured to ensure the consistency of the mode transitions of the home, and to ensure the correctness thereof.

The reasoning engine 16 is configured to activate the behaviors expected by the applicable rules.

Advantageously, the reasoning engine activates the expected behaviors according to a priority override mechanism (increasing from top to bottom).

This makes it possible to manage any overlapping of different rules, ensuring that implementation commands always result from the highest priority rules.

In particular, the priority override mechanism makes it possible to avoid implementation conflicts: the mechanism, due to the way it is constituted, ensures that each object is associated with one and only one implementation, thus avoiding the generation of conflicts (several implementations, resulting from different rules, associated with the same object).

Moreover, such mechanism ensures a higher priority to more specific rules, if any, i.e. rules with a higher priority: categories on base, moments on categories, user on previous ones, emergency on all.

According to a preferred embodiment, the applicable rales comprise the following rules, reported with increasing priority:

- basic rales 16a;

- category rales 16b;

- instant rales 16c;

- user rales 16d; - emergency rules 16e.

For example, the basic rules 16a may comprise: automatic change of seasonal home settings (transition of the home from summer to winter and vice versa); activation of automatic irrigation starting from current and expected weather conditions.

The category rules 16b may comprise, e.g., lighting, air quality, temperature, security, energy, irrigation.

The instant rules 16c may comprise, e.g., cinema time, rest time, party time.

The user rules 16d may comprise, e.g., punctual modification of the general rules, some examples may be the inhibition of punctual lighting of individual objects or categories.

The emergency rules 16e may comprise, e.g., securing the home in cases of emergency such as gas leaks or flooding.

The reasoning engine is configured to generate a set of actuation commands 17 or one (or several) user notification(s) 18 according to an evaluation of all the applicable rules 16a-16e.

Advantageously, the reasoning unit 14 also comprises a Machine Learning component 19 configured to create new knowledge and modify default behaviors starting from possible detected patterns.

In practice, therefore, unlike the rule systems present in literature, which require a“new version” of the rule set, and a restart of the reasoning process, every time one or more rules change, the automatic reasoner of the system 1 permits “codifying” the rules, i.e. the events that make it active and the actions that characterize it, by means of specific data structures. This makes it possible to change/adapt a set of rules and, therefore, the relative behavior of the automatic reasoner, without having to create and distribute a new version of the automatic reasoner or start the process again.

Usefully, each service is designed in itself in such a way as to reduce start times to the utmost.

In addition, thanks to the presence of data distribution and synchronization means 8 on all the nodes 2, each service can be performed in any of the nodes 2 of the network. This feature makes the system 1 extremely fault-tolerant.

A further advantage is that the location of the node services 9 and of the cluster services 10 makes it possible to ensure the privacy of users. All home-related information, whether usage data or multimedia information, is extracted and processed locally, and the processing results are maintained locally. The user may temporarily request access to such data only from the mobile application, subject to the privacy policies for all family members.

The sensor device 4 is selected out of: at least one camera, at least one motion radar, at least one temperature sensor, at least one humidity sensor, at least one light sensor, at least one sensor of CO2, CO or volatile gases, at least one microphone, at least one pressure sensor.

According to a preferred embodiment of the system 1, each of the nodes 2 comprises a plurality of sensor devices 4.

Preferably, each node 2 has the same set of sensor devices 4.

The data collected by the sensor devices 4 comprise at least one of either:

- detection of objects

- detection of animals

- detection of people

- identities of people.

For example, by collecting faces, clusters of all the people present are created. If the cluster is associated with a user, he/she will be recognized as“known”. More specifically, in this case, the term“cluster” means a set of techniques for the statistical analysis of multidimensional data, used for various purposes such as dimensionality reduction or unsupervised classification.

This information is extracted locally, appropriately filtered and compressed (to comply with the traffic constraints of the local wireless network), and processed in a node 2 elected to the function of automatic reasoner.

The automatic reasoner is configured to process the collected data to obtain at least the following semantically relevant information:

- presence of persons, or specific persons;

- environmental events; - social activities.

For example, the automatic reasoner is able to recognize, through analysis of the data produced in the system, some activities of interest such as the entry and exit of people from the rooms and the house, the home exit and return habits of the residents.

In particular, the automatic reasoner is configured for running in real time predefined rules of operation depending on said determined information.

By exploiting the distributed knowledge extracted from the entire house, such rules make it possible to provide the user with a series of smart behaviors adapted to improve comfort, lower the energy impact of the building and ensure security. For example: by learning when the inhabitants return home, the automatic reasoner turns the heating on; if the kW threshold is exceeded before the meter trips, it turns off the last load which has been turned on and which can be turned off; a resident is alerted if the system detects a stranger in a floor of the house where there are no other residents.

It has in practice been ascertained that the described invention achieves the intended objects.

In particular, the fact is underlined that, thanks to the local processing of the collected data, the system according to the invention is able to ensure the privacy of the users inasmuch as all the sensitive data (such as, e.g., videos) are processed locally and are not displayed on remote devices, except when the users are away from home.

Particular in this context, for example, are the set of algorithms and strategies necessary to generate the identity classifiers of individuals with only the information present in the distributed context of the house (i.e. using the visual information coming from all the rooms of the house), and to effectively distribute the output of these classifiers so as to minimize the traffic generated and ensure the privacy of users (i.e. without saving data directly related to the same, such as faces).

In addition, the system according to the invention allows the effective use of heterogeneous data. In fact, the use of distributed knowledge derived from the aggregation of heterogeneous data, and the consequent generation of data semantically more and more relevant in the domestic context, provides not only the possibility of further processing and refinement, but also determines a sort of high-level“language” for the description of the state of the house and for the expected behavior.

Furthermore, the use of a wireless mesh network makes it possible to install the system quickly and cost-effectively.

A further advantage is represented by the fact that the system is fault-tolerant.

In fact, the system consists of a plurality of devices and in case of failure of one of the devices (e.g. due to hardware problems) or in the event of one of the devices being unreachable (e.g. due to problems with the quality of the wireless communication means), the system nonetheless continues to operate correctly thanks to the clustering service.

Moreover, the system according to the invention is able to work correctly and effectively even in the absence of an Internet connection.