Scene generation using probabilistic programming languages
A system includes a memory device, and a processing device, operatively coupled to the memory device, to receive a set of input data including a roadgraph and a distribution of a plurality of scene types associated with a driving context, the roadgraph including an autonomous vehicle driving path, perform stochastic sampling based on the distribution of the plurality of scene types to obtain a set of synthetic scenes for the driving context, and train a machine learning model to identify, for each synthetic scene of the set of synthetic scenes, a respective modified autonomous vehicle driving path.
1 . A system comprising:
a memory device; and
a processing device, operatively coupled to the memory device, to:
receive a set of input data including a roadgraph and a first distribution of a plurality of scene types associated with a driving context, the roadgraph representing an autonomous vehicle driving path, and the first distribution of the plurality of scene types comprising the plurality of scene types each associated with a weight indicating a probability of an occurrence of a respective scene type;
generate a set of synthetic scenes for the driving context by performing stochastic sampling based on the first distribution of the plurality of scene types, wherein the processing device is to place a set of artifacts to define a synthetic scene within the roadgraph based on a second distribution of a variable associated with requirements for the driving context;
train a machine learning model to identify, for each synthetic scene of the set of synthetic scenes, a respective modified autonomous vehicle driving path to be navigated by an autonomous vehicle; and
cause an autonomous vehicle control system to modify, using the trained machine learning model, the autonomous driving path.
2 . The system of claim 1 , wherein each synthetic scene of the set of synthetic scenes is a synthetic construction zone.
3 . The system of claim 1 , wherein each respective modified autonomous vehicle path includes at least one of: a path shift, or a path merge into a second autonomous vehicle driving path represented by the roadgraph.
4 . The system of claim 1 , wherein the first distribution of the plurality of scene types comprises the plurality of scene types each associated with the weight indicating the probability of occurrence of the respective scene type under realistic driving conditions.
5 . The system of claim 1 , wherein the second distribution follows a probability density function.
6 . The system of claim 1 , wherein the stochastic sampling is performed by a probabilistic programming language module (PPLM) written in a probabilistic programming language and nested within a program written in a general purpose programming language.
7 . The system of claim 1 , wherein the set of synthetic scenes comprises a plurality of synthetic scenes, each synthetic scene of the plurality of synthetic scenes being associated with an intraclass variety.
8 . The system of claim 1 , wherein the processing device is further to:
generate a set of training input data comprising a set of data frames from the set of synthetic scenes; and
obtain a set of target output data for the set of training input data, wherein the machine learning model is trained using the set of training input data and the set of target output data.
9 . The system of claim 8 , wherein the set of target output data comprises at least one of: messages with injected markers or perception objects, or tensorflow examples.
10 . A method comprising:
receiving, by a processing device, a set of input data including a roadgraph and a first distribution of a plurality of scene types associated with a driving context, the roadgraph representing an autonomous vehicle driving path, and the first distribution of the plurality of scene types comprising the plurality of scene types each associated with a weight indicating a probability of an occurrence of a respective scene type;
generating, by the processing device, a set of synthetic scenes for the driving context by performing stochastic sampling based on the first distribution of the plurality of scene types, wherein the processing device is to place a set of artifacts to define a synthetic scene within the roadgraph based on a second distribution of a variable associated with requirements for the driving context;
training, by the processing device, a machine learning model to identify, for each synthetic scene of the set of synthetic scenes, a respective modified autonomous vehicle driving path to be navigated by an autonomous vehicle; and
causing an autonomous vehicle control system to modify, using the trained machine learning model, the autonomous driving path.
11 . The method of claim 10 , wherein each synthetic scene of the set of synthetic scenes is a synthetic construction zone.
12 . The method of claim 10 , wherein each respective modified autonomous vehicle path includes at least one of: a path shift, or a path merge into a second autonomous vehicle driving path represented by the roadgraph.
13 . The method of claim 10 , wherein the first distribution of the plurality of scene types comprises the plurality of scene types each associated with the weight indicating the probability of occurrence of the respective scene type under realistic driving conditions.
14 . The method of claim 10 , wherein the second distribution follows a probability density function.
15 . The method of claim 10 , wherein the stochastic sampling is performed by a probabilistic programming language module (PPLM) written in a probabilistic programming language and nested within a program written in a general purpose programming language.
16 . The method of claim 10 , wherein the set of synthetic scenes comprises a plurality of synthetic scenes, each synthetic scene of the plurality of synthetic scenes being associated with an intraclass variety.
17 . The method of claim 10 , further comprising:
generating, by the processing device, a set of training input data comprising a set of data frames from the set of synthetic scenes; and
obtaining a set of target output data for the set of training input data, wherein the machine learning model is trained using the set of training input data and the set of target output data.
18 . The method of claim 17 , wherein the set of target output data comprises at least one of: messages with injected markers or perception objects, or tensorflow examples.
19 . A non-transitory computer-readable storage medium having instructions stored thereon that, when executed by a processing device, cause the processing device to:
receive a set of input data including a roadgraph and a first distribution of a plurality of scene types associated with a driving context, the roadgraph representing an autonomous vehicle driving path, and the first distribution of the plurality of scene types comprising the plurality of scene types each associated with a weight indicating a probability of an occurrence of a respective scene type;
generate a set of synthetic scenes for the driving context by performing stochastic sampling based on the first distribution of the plurality of scene types, wherein the processing device is to place a set of artifacts to define a synthetic scene within the roadgraph based on a second distribution of a variable associated with requirements for the driving context;
train a machine learning model to identify, for each synthetic scene of the set of synthetic scenes, a respective modified autonomous vehicle driving path to be navigated by an autonomous vehicle; and
cause an autonomous vehicle control system to modify, using the trained machine learning model, the autonomous driving path.
20 . The non-transitory computer-readable storage medium of claim 19 , wherein each synthetic scene of the set of synthetic scenes is a synthetic construction zone.