Apparatus and method for generating a first agent, in particular for an interaction between the first agent and a second agent, and apparatus and method for training at least one model for generating the first agent

Abstract

Apparatus and method for training at least one model or for generating a first agent, in particular for an interaction between the first agent and a second agent, wherein a description of a behavior of the first agent, in particular in the interaction between the first agent and the second agent, is mapped onto the first representation using a first model (202) that is designed to map the description onto a first representation, wherein the first representation is mapped onto the second representation using a second model (208) that is designed to map the first representation onto a second representation, wherein the second representation is mapped onto the output variable (214) using a third model (212) that is designed to map the second representation onto an output variable (214) for influencing the behavior of the first agent,wherein the description is specified in natural language, in particular in text form or audio form, or in formal language or in digital graphic form, wherein the behavior of the first agent, in particular in the interaction between the first agent and the second agent, is specified as a function of the output variable (214).

Description

State of the art

[0001] The invention is based on a device and a method for generating a first agent, in particular for an interaction between the first agent and a second agent, and a device and a method for training at least one model for generating the first agent. Disclosure of the invention

[0002] A method for generating a first agent, in particular for interaction between the first agent and a second agent, provides that a description of a behavior of the first agent, in particular in the interaction between the first agent and the second agent, is mapped onto the first representation using a first model that is designed to map the description onto a first representation, wherein the first representation is mapped onto the second representation using a second model that is designed to map the first representation onto a second representation, wherein the second representation is mapped onto the output variable using a third model that is designed to map the second representation onto an output variable for influencing the behavior of the first agent, wherein the description is in natural language, in particular in text form or audio form,or is specified in formal language or in digital graphic form, whereby the behavior of the first agent, particularly in the interaction between the first agent and the second agent, is specified as a function of the output variable. The respective model can be a stochastic, probabilistic model, or a deterministic model.

[0003] For example, an anomaly in the behavior of the second agent in the interaction between the first agent and the second agent is detected depending on the interaction.

[0004] It can be provided that the output variable comprises a trajectory of the first agent and / or that the output variable comprises controller parameters for a controller of the first agent, wherein the behavior of the first agent is determined depending on a behavior of the controller in the first agent.

[0005] It can be provided that the first model comprises a pre-trained artificial neural network and / or the second model comprises a pre-trained artificial neural network and / or that the third model comprises a pre-trained artificial neural network.

[0006] The method for training at least one model for generating a first agent, in particular for an interaction between a first agent and a second agent, provides that a description of a behavior of the first agent, in particular in the interaction between the first agent and the second agent, is mapped onto the first representation using a first model that is designed to map the description onto a first representation, wherein the first representation is mapped onto the second representation using a second model that is designed to map the first representation onto a second representation, wherein the second representation is mapped onto the output variable using a third model that is designed to map the second representation onto an output variable, wherein the description is specified in natural language or in formal language or in digital graphic form,The description and a reference for the output variable are specified, the reference characterizing realistic real-world behavior of the first agent that matches the description, and the second model is trained based on a difference between the output variable and the reference. This trains the second model to output output variables that define the most physically realistic real-world behavior of the first agent that matches the description.

[0007] It may be provided that the first model comprises a pre-trained artificial neural network and / or that the third model comprises a pre-trained artificial neural network. This means that the training is based on already available models.

[0008] It can be planned that the first model and / or the third model remain unchanged during training. This allows the second model to be trained specifically. This requires less training data than if the second model were trained together with the first model and / or the second model.

[0009] It can be provided that the reference comprises a trajectory of the first agent and / or controller parameters for a controller of the first agent.

[0010] A device for generating interactions or for training at least one model or for training a first agent for an interaction between the first agent and a second agent comprises at least one processor and at least one memory, wherein the at least one processor is designed to execute instructions, upon execution of which by the at least one processor the device executes the method, wherein the at least one memory stores the instructions.

[0011] A data structure comprises at least one data field for a description of a behavior of a first agent, in particular in the interaction between the first agent and a second agent, in natural language or in formal language, wherein the data structure comprises at least one data field for a first representation of the description, wherein the data structure comprises at least one data field for a second representation of the description, wherein the data structure comprises at least one data field for an output variable for influencing the behavior of the first agent.

[0012] It can be provided that the data structure comprises at least one data field for a first model which is designed to map the description to the first representation, wherein the data structure comprises at least one data field for a second model which is designed to map the first representation to the second representation, and / or wherein the data structure comprises at least one data field for a third model which is designed to map the second representation to the output variable.

[0013] A computer program may be provided which comprises computer-executable instructions, the execution of which by the computer causes the method to run on the computer.

[0014] Further advantageous embodiments can be found in the following description and the drawing. The drawing shows: Fig. 1 a schematic representation of a device for machine learning or for generating a behavior of a first agent, Fig. 2 a schematic representation of models for machine learning or for generating the behavior of the first agent, Fig. 3 a schematic representation of an exemplary interaction, Fig. 4 a flowchart with steps of the method for generating a behavior of the first agent, Fig. 5 a flowchart showing steps of the machine learning procedure.

[0015] In Fig. 1 schematically illustrates a device 100 for machine learning or for generating a behavior of a first agent, in particular in an interaction between the first agent and a second agent.

[0016] The device 100 comprises at least one processor 102 and at least one memory 104.

[0017] The at least one processor 102 is configured to execute instructions, upon execution of which by the at least one processor 102, the device 100 executes a method for machine learning or for generating the behavior described below.

[0018] The at least one memory 104 stores the instructions. The at least one memory 104 comprises, for example, a non-volatile memory. The at least one memory 104 comprises, for example, a volatile memory.

[0019] In the example, the device 100 comprises an interface 106. The interface 106 is designed, for example, to receive a description of the behavior.

[0020] The interface 106 may be configured to capture the description in text form, in audio form, or in digital graphic form.

[0021] The interface 106 can be configured to request the description via an output in text or audio form. The device 100 is configured, for example, to request and capture the description in a dialog with a user.

[0022] For example, device 100 is configured to automatically query information required for the description and add it to the description. Device 100 is configured, for example, to generate the description when device 100 detects that the information required for the description has been recorded.

[0023] In Fig. 2 shows schematic models for machine learning or for generating behavior.

[0024] In the example, a first model 202 is configured to map an input variable 204 of the first model 202 to an output variable 206 of the first model 202. The first model 202 is, for example, an encoder, or a sequential and / or autoregressive encoder, or a transformer architecture.

[0025] In the example, a second model 208 is configured to map the output variable 206 of the first model 202 to an output variable 210 of the second model 208. The second model 208 is, for example, a translator that translates the output variable 206 of the first model 202 into the output variable 210 of the second model 208, ie, an input variable for the third model 212.

[0026] In the example, a third model 212 is configured to map the output variable 210 of the second model 208 to an output variable 214 of the third model 212. The third model 212 is, for example, a decoder.

[0027] In the example, the input variable 204 of the first model 202 comprises the description in natural language or formal language or in digital graphic form. The output variable 206 of the first model 202 comprises a first representation in the example. The first representation is, for example, a first embedding or a first sequence of tokens. The output variable 210 of the second model 208 comprises a second representation in the example. The second representation is, for example, a second embedding or a second sequence of tokens. The output variable 214 of the third model 212 defines the behavior.

[0028] The first model 202 comprises, for example, an artificial neural network. The second model 208 comprises, for example, an artificial neural network. The third model 212 comprises, for example, an artificial neural network.

[0029] The device 100 includes, for example, the models. The device 100 is configured, for example, to receive the description via the interface 106 and to determine the output variable for the description using the models.

[0030] It can be provided that the device 100 is configured to conduct tests. The device 100 is configured, for example, to check the behavior of the second agent in the interaction with the first agent in a test depending on the interaction. The device 106 is configured, for example, to detect an anomaly in the interaction during the test and, as a result of the test, to output the presence of the anomaly via the interface 106 or to output via the interface 106 that no anomaly was detected in the test.

[0031] The interaction is not limited to one agent to be tested or one agent that can be moved synthetically during the interaction. Multiple agents can be provided that are to be tested with the interaction in the test. Multiple agents that can move synthetically during the interaction in the test can be provided. The description describes, for example, the behavior of the respective agent to be moved synthetically. The behavior of the respective synthetically movable agents is defined, for example, by the output variable 214.

[0032] In Fig. As an example of the interaction, an exemplary scenario 300 with a first vehicle 302, as an example of a first agent, and a second vehicle 304, as an example of a second agent, is depicted in FIG. The scenario 300 includes a trajectory 306 of the first vehicle 302 and a trajectory 308 of the second vehicle 304.

[0033] Scenario 300 is illustrated for the following exemplary description in natural language: Motorway entrance scenario where a first vehicle moving on the motorway entrance must merge after a second vehicle moving on the motorway.

[0034] In the example, the first vehicle 302 includes at least one sensor for acquiring sensor data that characterizes an environment of the vehicle 302.

[0035] In the example, the first vehicle 302 includes a controller for regulating the behavior of the first vehicle 302 depending on the sensor data. The controller is parameterized, for example, by controller parameters.

[0036] In the test, for example, the second vehicle 304 is checked in its interaction with the first vehicle 302.

[0037] The graphic representation of scenario 300 represents trajectory 306 as an exemplary description of the behavior of first vehicle 302 in digital graphic form. The graphic representation is exemplary. Device 100 can be configured to use scenario 300 in a formal language that can be processed automatically during testing, for example, by a test bench or a simulation environment in which the test is performed.

[0038] It may be provided that the description provided as input variable 204 of the first model 202 is supplemented by information in natural language on the framework conditions. These framework conditions define, for example, in natural language a geometry of the highway, Weather conditions such as dry, windy, rain, snow, black ice, visibility conditions such as foggy, dark, bright, Traffic rules, such as speed limits and overtaking bans.

[0039] Scenario 300, for example, includes a description of a map that includes the boundary conditions. The map description is provided, for example, in a formal language that can be processed automatically by the test bench or simulation environment.

[0040] In Fig. Figure 4 shows a flowchart with steps of a procedure.

[0041] In one example, the agents are road users. The first agent is, for example, the first vehicle 302. The second agent is, for example, the second vehicle 304.

[0042] In a robotics example, the first agent is a robot and the second agent is a human model.

[0043] The procedure is based on a given description.

[0044] The description is provided, for example, in natural language, particularly in text or audio form, or in formal language, or in digital graphic form. A user enters the description, for example, in text form, or speaks the description in audio form. A user draws the description in digital graphic form, e.g., in the form of a sketch.

[0045] The description is requested, for example, through text or audio output. The description is requested and recorded, for example, in a dialog with a user.

[0046] The information required for the description is automatically queried and added to the description.

[0047] For example, the description is only generated when it is recognized that the information required for the description has been recorded.

[0048] For example, a description of an interaction that is rare in the real world is captured, in particular using language. An example of a rare interaction is an interaction in which the second agent is an autonomous vehicle, where the first agent is another vehicle that unexpectedly drives in front of the autonomous vehicle from a parking space that is not visible to the autonomous vehicle. An example of a rare interaction is an interaction with aggressive driving by an agent, e.g. when the agent merges into traffic with other agents on a motorway. An example of a rare interaction is an interaction under extreme weather conditions in which the first agent and / or the second agent are located. An example of a rare interaction is an interaction under very rare environmental conditions, e.g. on a day on which the first agent represents a person who is in costume, such as on Halloween or Mardi Gras.Carnival. In this context, rare means that the interaction occurs very rarely in training data collected in the real world.

[0049] The method is based on the first model 202, the second model 208, and the third model 212. The first model 202, the second model 208, and the third model 212 are predefined in the example method. For example, the respective artificial neural network is pretrained.

[0050] The first model 202 is, for example, a model pre-trained for text inputs and / or audio inputs, which is designed to map the description in text form or in audio form to the first representation. Using the pre-trained first model 202 for this purpose enables greater data efficiency compared to a single model that is trained to map the description directly to the output variable 214.

[0051] The behavior of the first agent is generated, for example, in a specific domain, e.g., transportation or robotics. The first model 202 and / or the third model 212 are pre-trained, for example, in a different domain or without content specialization on the domain in which the agent's behavior is generated. The domain in which the first model 202 and / or the third model 212 are pre-trained is a different domain, for example, a more general domain, than the domain in which the behavior of the first agent is generated.

[0052] The pre-trained first model 202 and / or the pre-trained third model 212 enables knowledge transfer, in particular from a domain in which the first model 202 or the third model 212 is pre-trained, to the domain in which the first agent is generated. For example, the respective pre-trained model can comprise general knowledge about a behavior of the first agent. For example, a behavior of a pedestrian represented by the first agent can be better generated based on the general knowledge if the respective pre-trained model is pre-trained with more data comprising human behavior. In a subsequent training of the respective pre-trained model to generate the first agent, less data with pedestrian behavior is then required to generate the behavior of the first agent in a comparably good manner, as is otherwise only possible with more data with pedestrian behavior.

[0053] The method includes a step 402.

[0054] In step 402, the given description is mapped to the first representation using the first model 202.

[0055] The method includes a step 404.

[0056] In step 404, the first representation is mapped to the second representation using the second model 208.

[0057] The method includes a step 406.

[0058] In step 406, the second representation is mapped to the output variable 214 using the third model 212.

[0059] The method may include further steps for training and / or testing the behavior of the second agent.

[0060] Steps 402 to 406 are repeated, for example, to generate training data comprising output variables 214 for training and / or testing the behavior of the second agent. For example, a plurality of output variables 214 are determined from a catalog of descriptions and added to the training data.

[0061] The catalog includes, for example, a description of interactions in a domain. An example of a domain description is the operational design domain described in Koopman, P., Osyk, B., Weast, J. (2019). Autonomous Vehicles Meet the Physical World: RSS, Variability, Uncertainty, and Proving Safety. In: Romanovsky, A., Troubitsyna, E., Bitsch, F. (eds) Computer Safety, Reliability, and Security. SAFECOMP 2019. Lecture Notes in Computer Science(), vol 11698. Springer, Cham. https: / / doi.org / 10.1007 / 978-3-030-26601-1_17.

[0062] For example, the method for training includes a step 408.

[0063] In step 408, a behavior of the second agent is trained in the interaction between the first agent and the second agent

[0064] The behavior of the first agent is determined, for example, by the trajectory specified for the first agent in the scenario.

[0065] The first agent may include a controller configured to determine the first agent's behavior depending on the scenario. The first agent's behavior is determined, for example, by the controller in the first agent.

[0066] For example, the second agent is trained using reinforcement learning, i.e., reenforcement learning. For example, the second agent is rewarded if it does not collide with the first agent moving along the given trajectory, and is not rewarded otherwise.

[0067] The second agent can comprise a sensor and a controller configured to determine the behavior of the second agent based on information about the first agent measured by the sensor. The behavior of the second controller is determined, for example, by the controller in the second agent. For example, the controller of the second agent is trained using reinforcement learning.

[0068] For example, the method for testing includes a step 410.

[0069] In step 410, the behavior of the second agent is monitored. It may be provided that the behavior of the first agent is monitored, or that the behavior of the first agent generated with the models is used as a basis for testing.

[0070] The behavior of the first agent is determined, for example, by the trajectory or the controller in the first agent.

[0071] For example, the method for testing includes a step 412.

[0072] In step 412, an anomaly in the behavior of the second agent is detected depending on the interaction of the second agent with the first agent. For example, the anomaly in the behavior of the second agent is detected depending on the interaction.

[0073] For example, the anomaly in the behavior of the second agent is detected when the first agent collides with the second agent.

[0074] The method can be used to generate multiple agents, including their interaction behavior with each other and with potential third parties. It can be used to test the generated agents' interactions with one or more external agents in a real-world test environment or in a simulation.

[0075] The test itself is carried out, for example, in the real test environment or in the simulation in which the agents interact. Simulation is particularly preferable for tests that could lead to collisions, in order to avoid endangering human lives or to prevent the destruction of real agents.

[0076] For example, a description of a merging scenario in which two agents are driving with little space between them in the area of ​​an on-ramp on a highway generates a behavior of two agents driving on the highway according to the described merging scenario.

[0077] For example, an external agent whose behavior is not generated by the description but is controlled by an externally specified controller is tested to see whether this external agent still threads in without accidents.

[0078] In addition, the test can be performed for the domain in which the first agent is created and for further descriptions specified by a user.

[0079] It may be planned to conduct further tests, e.g., with interactions captured in the real world. It may be planned to randomly select the interactions captured in the real world from a set of predefined interactions captured in the real world.

[0080] In Fig. 5 is a flowchart showing steps of a method for training at least one of the models.

[0081] The training method is based on the first model 202, the second model 208, and the third model 212. In the example, the first model 202 and the third model 212 are predefined in the method for training at least one of the models. For example, the respective neural network of the first model 202 and the third model 212 is pretrained.

[0082] Pre-trained models result in knowledge transfer and / or higher data efficiency due to the information stored in pre-training and the density of information contained in language compared to models learned only on the domain in which the first agent is generated.

[0083] The method for training at least one of the models is based on training data. In the example, the training data comprises training data points, each of which has a predefined description and a reference associated with the predefined description. Furthermore, the method can be based on pre-trained models.

[0084] The method for training at least one of the models comprises a step 502.

[0085] In step 502, the description and reference from a training data point are specified.

[0086] In step 504, for a training data point, the given description is mapped to the first representation using the first model 202.

[0087] The method for training at least one of the models comprises a step 506.

[0088] In step 506, for the training data point, the first representation is mapped to the second representation using the second model 208.

[0089] The method for training at least one of the models comprises a step 508.

[0090] In step 508, the second representation for the training data point is mapped to the output variable 214 using the third model 212.

[0091] The reference includes, for example, an output variable for a realistic, real-world behavior that fits the description. The output variable includes, for example, a trajectory that defines the behavior of the first agent, a controller parameter for the controller of the first agent, and / or a map that defines boundary conditions for the behavior of the first agent.

[0092] In the example, steps 502 to 508 are performed for the training data points from the training data.

[0093] The method for training at least one of the models comprises a step 510.

[0094] In step 510, the second model 208 is trained depending on a difference between the output variable 214 and the reference.

[0095] In the example, the second model 208 is trained based on an objective function that includes the respective difference determined for the training data points. For example, the second model 208 is determined with which an objective function that depends on a sum of the differences is as small as possible, in particular minimal.

[0096] For example, parameters of the neural network comprising the second model 208 are determined using a gradient descent method depending on the differences.

[0097] In the example, the first model 202 and the third model 212 remain unchanged during the training of the second model 202. It may be planned to also train the first model 202 and / or the third model 212 during training.

[0098] The output variables of the models are summarized in vectors, for example. For example, a trajectory or the controller parameters in the scenario are described by a vector output by the third model 212. The map, for example, is described by a vector output by the third model 212, which includes the parameters of the formal language in which the map is described.

[0099] The first model 202, the second model 208, and / or the third model 212 may be a stochastic, probabilistic model, or a deterministic model.

[0100] The first model 202, the second model 208, and the third model 212 are configured in an example of a mapping in the reverse direction. The mapping in the reverse direction enables the interpretability of behavior because the models are correlated with language. This means that the third model 212 is configured to map the output variable 214 of the third model 212 to the output variable 210 of the second model 208. The second model 208 is configured to map the output variable 210 of the second model 208 to the output variable 206 of the first model 202. The first model 202 is configured to map the output variable 206 of the first model 202 to the input variable 204 of the first model 202.

[0101] For mapping in the reverse direction, it can be provided that frames of a video are used successively for the output variable 214 of the third model 212, with the video being mapped to a textual description of the video's content. This means that the appropriate textual description is generated from a behavioral video.

[0102] A method for explaining a behavior of the second agent, in particular in the interaction with the first agent, provides that the output variable 214 of the third model 212 includes the behavior to be explained.

[0103] The method for explanation provides that the output variable 214 of the third model 212 is mapped to the output variable 210 of the second model 108.

[0104] The method for explanation provides that the output variable 210 of the second model 208 is mapped to the output variable 206 of the first model 202.

[0105] The explanation method provides that the output variable 206 of the first model 202 is mapped to the input variable 204 of the first model 202. The input variable 204 of the first model 202 includes the description of the behavior to be explained, in particular the interaction to be explained between the first agent and the second agent. QUOTES CONTAINED IN THE DESCRIPTION

[0000] This list of documents submitted by the applicant was generated automatically and is included solely for the convenience of the reader. This list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions. Cited non-patent literature

[0000] Koopman, P., Osyk, B., Weast, J. (2019). Autonomous Vehicles Meet the Physical World: RSS, Variability, Uncertainty, and Proving Safety. In: Romanovsky, A., Troubitsyna, E., Bitsch, F. (eds) Computer Safety, Reliability, and Security. SAFECOMP 2019. Lecture Notes in Computer Science(), vol 11698. Springer, Cham. https: / / doi.org / 10.1007 / 978-3-030-26601-1_17

[0061]

Claims

[1] Method for generating a first agent (302), in particular for an interaction (300) between the first agent (302) and a second agent (304), characterized bythat a description of a behavior of the first agent (302), in particular in the interaction (300) between the first agent (302) and the second agent (304), is mapped (402) onto the first representation using a first model (202) that is designed to map the description onto a first representation, wherein the first representation is mapped (404) onto the second representation using a second model (208) that is designed to map the first representation onto a second representation, wherein the second representation is mapped (406) onto the output variable (214) using a third model (212) that is designed to map the second representation onto an output variable (214) for influencing the behavior of the first agent (302), wherein the description is specified in natural language, in particular in text form or audio form, or in formal language or in digital graphic form,wherein the behavior of the first agent (302), in particular in the interaction (300) between the first agent (302) and the second agent (304), is predetermined (408) depending on the output variable (214). [2] Method according to claim 1, characterized by that an anomaly in the behavior of the second agent (304) in the interaction between the first agent (302) and the second agent (304) is detected (412) depending on the interaction (300). [3] Method according to claim 1 or 2, characterized by that the output variable (214) comprises a trajectory of the first agent (302) and / or that the output variable (214) comprises controller parameters for a controller of the first agent (302), wherein the behavior of the first agent (302) is determined (410) depending on a behavior of the controller in the first agent (302). [4] Method according to one of the preceding claims, characterized bythat the first model (202) comprises a pre-trained artificial neural network and / or the second model (208) comprises a pre-trained artificial neural network and / or that the third model (212) comprises a pre-trained artificial neural network. [5] Method for training at least one model for generating a first agent (302), in particular for an interaction (300) between a first agent (302) and a second agent (304), characterized bythat a description of a behavior of the first agent (302), in particular in the interaction (300) between the first agent (302) and the second agent (304), is mapped (504) onto the first representation using a first model (202) that is designed to map the description onto a first representation, wherein the first representation is mapped (506) onto the second representation using a second model (208) that is designed to map the first representation onto a second representation, wherein the second representation is mapped (508) onto the output variable (214) using a third model (212) that is designed to map the second representation onto an output variable (214) for influencing the behavior of the first agent (302), wherein the description is specified in natural language or in formal language or in digital graphic form,wherein the description and a reference for the output variable (214) are specified (502), wherein the reference characterizes a behavior of the first agent (302) that is realistic in the real world and matches the description, and wherein the second model is trained (510) depending on a difference between the output variable (214) and the reference. [6] Method according to claim 5, characterized by that the first model (202) comprises a pre-trained artificial neural network and / or that the third model (212) comprises a pre-trained artificial neural network. [7] Method according to claim 6, characterized by that the first model (202) and / or the third model (212) remain unchanged during training (510). [8] Method according to one of claims 5 to 7, characterized by that the reference comprises a trajectory of the first agent (302) and / or controller parameters for a controller of the first agent (302). [9] Device (100) for generating interactions or for training at least one model or for training a first agent for an interaction (300) between the first agent (302) and a second agent (304), characterized by that the device (100) comprises at least one processor (102) and at least one memory (104), wherein the at least one processor (102) is designed to execute instructions, upon execution of which by the at least one processor (102) the device (100) carries out the method according to one of claims 1 to 8, wherein the at least one memory (104) stores the instructions. [10] Data structure, characterized bythat the data structure comprises at least one data field for a description of a behavior of a first agent (302), in particular in the interaction between the first agent (302) and a second agent (304), in natural language or in formal language, wherein the data structure comprises at least one data field for a first representation of the description, wherein the data structure comprises at least one data field for a second representation of the description, wherein the data structure comprises at least one data field for an output variable (214) for influencing the behavior of the first agent (302). [11] Data structure according to claim 10, characterized byin that the data structure comprises at least one data field for a first model (202) which is designed to map the description to the first representation, wherein the data structure comprises at least one data field for a second model (208) which is designed to map the first representation to the second representation, and / or wherein the data structure comprises at least one data field for a third model (212) which is designed to map the second representation to the output variable (214). [12] Computer program, characterized by that the computer program comprises computer-executable instructions, the execution of which by the computer causes the method according to one of claims 1 to 8 to run on the computer.

Images