Automated generation and refinement of variation parameters for simulation scenarios
Disclosed herein are system, method, and computer program product embodiments for generating and refining simulation scenarios. For example, the method includes generating multiple base scenarios, each including one or more constant and one or more variable parameters. For each of the base scenarios, the method includes generating scenario variations, each of which is associated with a unique combination of values assigned to its base scenario's parameters. The method further includes determining a system boundary in a parameter space defined by the variable parameters, wherein the system boundary divides the parameter space into a region including successful scenario variations and a region including unsuccessful scenario variations, and generating additional scenario variations within a threshold distance of the system boundary. The method further includes simulating operation of an autonomous vehicle (AV) using one or more generated scenario variations.
1 . A method for providing a computer simulation of an autonomous vehicle, comprising:
generating, by a computing device, a plurality of base scenarios, each of the plurality of base scenarios is executable by a simulator and comprises one or more constant parameters and one or more variable parameters defining a configuration of a simulation scenario that is an artificial representation of a real-world occurrence for the autonomous vehicle, wherein the one or more variable parameters comprises a parameter value for a circuit element of the autonomous vehicle;
for each of the plurality of base scenarios, performing the following operations by the computing device:
generating a plurality of scenario variations for each base scenario of the plurality of base scenarios, each said scenario variation specifying a simulation scene created using a unique combination of values assigned to that base scenario's one or more variable parameters,
defining a system boundary line in a parameter space for said base scenario that is defined by all possible values of the base scenario's one or more variable parameters, wherein the system boundary line divides the parameter space into a first region including successful scenario variations and a second region including unsuccessful scenario variations, and
generating additional scenario variations by resampling data points in the parameter space from a region within a threshold distance of the system boundary line and residing on a particular side of the system boundary line; and
computer simulating, by the simulator, operation of the autonomous vehicle in different conditions and environments by applying a plurality of scenarios to an autonomous vehicle system model, the plurality of scenarios comprising a subset of the plurality of base scenarios and one or more generated scenario variations.
2 . The method of claim 1 , further comprising generating a global system boundary in a design space, wherein the global system boundary divides the design space into a first region including base scenarios having more than a threshold number of successful scenario variations and a second region including base scenarios having fewer than a threshold number of successful scenario variations.
3 . The method of claim 2 , further comprising generating additional base scenarios within a threshold distance of the global system boundary.
4 . The method of claim 3 , further comprising prioritizing the plurality of base scenarios and the additional base scenarios based on a distance of each of the plurality of base scenarios and each of the additional base scenarios from the global system boundary.
5 . The method of claim 2 , further comprising removing a first base scenario as redundant when the first base scenario is determined to be within a threshold distance of a second base scenario in the design space, the threshold being determined based on a distance from the global system boundary.
6 . The method of claim 2 , wherein generating the base scenarios comprises generating base scenarios to include parameters that are uniformly sampled from the design space.
7 . The method of claim 1 , further comprising, for each of the plurality of base scenarios, prioritizing the plurality of scenario variations for that base scenario and the additional scenario variations based on a distance of each of the plurality of scenario variations and each of the additional scenario variations from the system boundary.
8 . The method of claim 7 , wherein simulating the operation of the AV using the plurality of scenarios comprises selecting, based on the prioritizing of the plurality of base scenarios and the prioritizing of the plurality of scenario variations, the subset of the plurality of base scenarios and a subset of scenario variations associated with each of the base scenarios in the plurality of base scenarios.
9 . The method of claim 1 , further comprising removing a first scenario variation as redundant when the first scenario variation is determined to be within a threshold distance of a second scenario variation in the parameter space, the threshold being determined based on a distance from the system boundary.
10 . The method of claim 1 , further comprising storing each base scenario and corresponding scenario variations as a scenario data object in a data store.
11 . The method of claim 1 , further comprising detecting a triggering event indictive of whether or not new scenarios need to be generated, the triggering event selected from at least one of the following: detection of change in a system boundary that is greater than a threshold, detection of change in a global system boundary that is greater than a threshold, a threshold amount of time since a previous scenario generation, receipt of a user instruction, initiation of a scenario refresh cycle, or an AV system change that requires new scenario generation.
12 . The vehicle motion planning model training system of claim 1 , wherein the programming instructions are further configured to cause the processor to train the vehicle motion planning model by prioritizing the plurality of base scenarios and the additional base scenarios based on a distance of each of the plurality of base scenarios and each of the additional base scenarios from the global system boundary.
13 . A vehicle motion planning model training system, comprising:
a processor;
a data store containing a plurality of simulation scenarios; and
a memory that stores programming instructions that are configured to cause the processor to train a vehicle motion planning model by:
generating a plurality of base scenarios, each of the plurality of base scenarios is executable by a simulator and comprises one or more constant parameters and one or more variable parameters, defining a configuration of a simulation scenario that is an artificial representation of a real-world occurrence for an autonomous vehicle, wherein the one or more variable parameters comprises a parameter value for a circuit element of the autonomous vehicle;
for each of the plurality of base scenarios:
generating a plurality of scenario variations for each base scenario of the plurality of base scenarios, each said scenario variation specifying a simulation scene created using a unique combination of values assigned to that base scenario's one or more variable parameters,
defining a system boundary line in a parameter space for said base scenario that is defined by all possible values of the base scenario's one or more variable parameters, wherein the system boundary line divides the parameter space into a first region including successful scenario variations and a second region including unsuccessful scenario variations, and
generating additional scenario variations by resampling data points in the parameter space from a region within a threshold distance of the system boundary line and residing on a particular side of the system boundary line; and
computer simulating operation of the autonomous vehicle (AV) in different conditional and environments by applying a plurality of scenarios to the vehicle motion planning model, the plurality of scenarios comprising a subset of the plurality of base scenarios and one or more generated scenario variations.
14 . The vehicle motion planning model training system of claim 13 , wherein the programming instructions are further configured to cause the processor to train the vehicle motion planning model by generating a global system boundary in a design space, wherein the global system boundary divides the design space into a first region including base scenarios having more than a threshold number of successful scenario variations and a second region including base scenarios having fewer than a threshold number of successful scenario variations.
15 . The vehicle motion planning model training system of claim 14 , wherein the programming instructions are further configured to cause the processor to train the vehicle motion planning model by generating additional base scenarios within a threshold distance of the global system boundary.
16 . A computer program product comprising:
a memory that stores programming instructions that are configured to cause a processor to train a vehicle motion planning model by:
generating a plurality of base scenarios, each of the plurality of base scenarios is executable by a simulator and comprises one or more constant parameters and one or more variable parameters defining a configuration of a simulation scenario that is an artificial representation of a real-world occurrence for the autonomous vehicle, wherein the one or more variable parameters comprises a parameter value for a circuit element of the autonomous vehicle;
for each of the plurality of base scenarios:
generating a plurality of scenario variations for each base scenario of the plurality of base scenarios, each said scenario variation specifying a simulation scene created using a unique combination of values assigned to that base scenario's one or more variable parameters,
defining a system boundary line in a parameter space for said base scenario that is defined by all possible values of the base scenario's one or more variable parameters, wherein the system boundary line divides the parameter space into a first region including successful scenario variations and a second region including unsuccessful scenario variations, and
generating additional scenario variations by resampling data points in the parameter space from a region within a threshold distance of the system boundary line and residing on a particular side of the system boundary line; and
computer simulating operation of the autonomous vehicle in different conditions and environments by applying a plurality of scenarios to the vehicle motion planning model, the plurality of scenarios comprising a subset of the plurality of base scenarios and one or more generated scenario variations.
17 . The computer program product of claim 16 , wherein the programming instructions are further configured to cause the processor to train the vehicle motion planning model by generating a global system boundary in a design space, wherein the global system boundary divides the design space into a first region including base scenarios having more than a threshold number of successful scenario variations and a second region including base scenarios having fewer than a threshold number of successful scenario variations.
18 . The computer program product of claim 17 , wherein the programming instructions are further configured to cause the processor to train the vehicle motion planning model by generating additional base scenarios within a threshold distance of the global system boundary.
19 . The computer program product of claim 18 , wherein the programming instructions are further configured to cause the processor to train the vehicle motion planning model by prioritizing the plurality of base scenarios and the additional base scenarios based on a distance of each of the plurality of base scenarios and each of the additional base scenarios from the global system boundary.
20 . The computer program product of claim 16 , wherein the programming instructions are further configured to cause the processor to train the vehicle motion planning model by, for each of the plurality of base scenarios, prioritizing the plurality of scenario variations for that base scenario and the additional scenario variations based on a distance of each of the plurality of scenario variations and each of the additional scenario variations from the system boundary.