Validating autonomous vehicle simulation scenarios
Validating a simulation scenario for use in training a machine learning model for an autonomous vehicle includes determining a simulation scenario; executing a simulation based on a simulation scenario; monitoring execution of the simulation and receiving messages from the execution of the simulation; determining, by a first simulation monitor, whether the messages satisfy a first incident; and validating the simulation scenario responsive to the messages satisfying the first incident to produce validated simulation data. A system and method may also include determining, by a first simulation validator, whether the messages satisfy a condition; and validating the simulation scenario responsive to the messages satisfying the condition to produce validated simulation data.
1 . A method of validating a simulation scenario to be used in training a machine learning model of an autonomous vehicle, the method comprising:
executing a simulation to simulate a behavior of the autonomous vehicle based on the simulation scenario, simulation data in the simulation scenario being a training example to be used in training the machine learning model prior to its deployment in the autonomous vehicle;
monitoring the execution of the simulation and receiving a first message including state information of the autonomous vehicle based on a simulated behavior of the autonomous vehicle during the execution of the simulation;
determining, by a first simulation validator, whether the first message including the state information of the autonomous vehicle satisfies a first condition, the first condition evaluating a first aspect of the simulated behavior of the autonomous vehicle against a first threshold;
responsive to determining that the first message including the state information of the autonomous vehicle satisfies the first condition, signaling a success in the execution of the simulation and validating the simulated behavior of the autonomous vehicle in the simulation scenario as being exclusive of the first aspect of the simulated behavior of the autonomous vehicle that may unwantedly bias the machine learning model;
generating a training instance including the validated simulated behavior of the autonomous vehicle in the simulation scenario; and
training the machine learning model using the training instance including the validated simulated behavior of the autonomous vehicle in the simulation scenario.
2 . The method of claim 1 , wherein the execution of the simulation generates a simulated output of the simulation scenario, and the method further comprises:
providing the validated simulated behavior of the autonomous vehicle in the simulation scenario as a training input to the machine learning model to generate a predicted output of the machine learning model; and
updating one or more weights in the machine learning model based on a difference between the predicted output and the simulated output of the simulation scenario.
3 . The method of claim 1 , further comprising signaling termination of the execution of the simulation in response to determining that the first message does not satisfy the first condition.
4 . The method of claim 1 , further comprising signaling one or more of:
a failure in the execution of the simulation in response to determining the first message does not satisfy the first condition; and
an advisory message about the execution of the simulation.
5 . The method of claim 1 , wherein the simulation scenario is a three-dimensional virtual scene simulating an encounter between the autonomous vehicle and an entity in a surrounding environment of the autonomous vehicle.
6 . The method of claim 1 , wherein the first simulation validator comprises one from a group of a timer validator, a speed validator, a distance validator, a collision validator, a passenger experience validator, a region validator, a lateral offset validator, a lateral wiggle validator, a traffic light validator, and a reckless driving validator.
7 . The method of claim 1 , wherein monitoring the execution of the simulation includes receiving a second message from the execution of the simulation, and the method further comprises:
determining, by a second simulation validator, whether the second message satisfies a second condition, the second condition evaluating a second aspect of the simulated behavior of the autonomous vehicle against a second threshold; and
wherein the validating the simulation data in the simulation scenario is responsive to a logical combination of the first message satisfying the first condition and the second message satisfying the second condition.
8 . The method of claim 7 , wherein the first simulation validator and the second simulation validator are a same type.
9 . The method of claim 7 , further comprising:
generating metrics associated with the first simulation validator and the second simulation validator; and
storing the metrics in a database.
10 . The method of claim 1 , further comprising:
receiving a plurality of messages from the execution of the simulation representing time series data;
buffering the plurality of messages for the first simulation validator; and
determining, by the first simulation validator, whether the plurality of messages satisfies the first condition.
11 . A system comprising one or more processors and memory operably coupled with the one or more processors, wherein the memory stores instructions that, in response to execution of the instructions by the one or more processors, cause the one or more processors to perform operations including:
executing a simulation to simulate a behavior of an autonomous vehicle based on a simulation scenario, simulation data in the simulation scenario being a training example to be used in training a machine learning model of the autonomous vehicle prior to its deployment in the autonomous vehicle;
monitoring the execution of the simulation and receiving a first message including state information of the autonomous vehicle based on a simulated behavior of the autonomous vehicle during the execution of the simulation;
determining, by a first simulation validator, whether the first message including the state information of the autonomous vehicle satisfies a first condition, the first condition evaluating a first aspect of the simulated behavior of the autonomous vehicle against a first threshold;
responsive to determining that the first message including the state information of the autonomous vehicle satisfies the first condition, signaling a success in the execution of the simulation and validating the simulated behavior of the autonomous vehicle in the simulation scenario as being exclusive of the first aspect of the simulated behavior of the autonomous vehicle that may unwantedly bias the machine learning model;
generating a training instance including the validated simulated behavior of the autonomous vehicle in the simulation scenario; and
training the machine learning model using the training instance including the validated simulated behavior of the autonomous vehicle in the simulation scenario.
12 . The system of claim 11 , wherein the execution of the simulation generates a simulated output of the simulation scenario and wherein the operations further comprise:
providing the validated simulated behavior of the autonomous vehicle in the simulation scenario as a training input to the machine learning model to generate a predicted output of the machine learning model; and
updating one or more weights in the machine learning model based on a difference between the predicted output and the simulated output of the simulation scenario.
13 . The system of claim 11 , wherein the operations further comprise signaling termination of the execution of the simulation in response to determining that the first message does not satisfy the first condition.
14 . The system of claim 11 , wherein the operations further comprise signaling one or more of:
a failure in the execution of the simulation in response to determining the first message does not satisfy the first condition; and
an advisory message about the execution of the simulation.
15 . The system of claim 11 , wherein the simulation scenario is a three-dimensional virtual scene simulating an encounter between the autonomous vehicle and an entity in a surrounding environment of the autonomous vehicle.
16 . The system of claim 11 , wherein the first simulation validator comprises one from a group of a timer validator, a speed validator, a distance validator, a collision validator, a passenger experience validator, a region validator, a lateral offset validator, a lateral wiggle validator, a traffic light validator, and a reckless driving validator.
17 . The system of claim 11 , wherein monitoring the execution of the simulation includes receiving a second message from the execution of the simulation and wherein the operations further comprise:
determining, by a second simulation validator, whether the second message satisfies a second condition, the second condition evaluating a second aspect of the simulated behavior of the autonomous vehicle against a second threshold; and
wherein the validating the simulation data in the simulation scenario is responsive to a logical combination of the first message satisfying the first condition and the second message satisfying the second condition.
18 . The system of claim 17 , wherein the operations further comprise:
generating metrics associated with the first simulation validator and the second simulation validator; and
storing the metrics in a database.
19 . The system of claim 11 , wherein the operations further comprise:
receiving a plurality of messages from the execution of the simulation representing time series data;
buffering the plurality of messages for the first simulation validator; and
determining, by the first simulation validator, whether the plurality of messages satisfies the first condition.
20 . A non-transitory computer readable storage medium storing computer instructions executable by one or more processors to perform a method of validating a simulation scenario to be used in training a machine learning model of an autonomous vehicle, the method comprising:
executing a simulation to simulate a behavior of the autonomous vehicle based on the simulation scenario, simulation data in the simulation scenario being a training example to be used in training the machine learning model prior to its deployment in the autonomous vehicle;
monitoring the execution of the simulation and receiving a first message including state information of the autonomous vehicle based on a simulated behavior of the autonomous vehicle during the execution of the simulation;
determining, by a first simulation validator, whether the first message including the state information of the autonomous vehicle satisfies a first condition, the first condition evaluating a first aspect of the simulated behavior of the autonomous vehicle against a first threshold;
responsive to determining that the first message including the state information of the autonomous vehicle satisfies the first condition, signaling a success in the execution of the simulation and validating the simulated behavior of the autonomous vehicle in the simulation scenario as being exclusive of the first aspect of the simulated behavior of the autonomous vehicle that may unwantedly bias the machine learning model;
generating a training instance including the validated simulated behavior of the autonomous vehicle in the simulation scenario; and
training the machine learning model using the training instance including the validated simulated behavior of the autonomous vehicle in the simulation scenario.