Generating autonomous vehicle simulation data from logged data
Logged data from an autonomous vehicle is processed to generate augmented data. The augmented data describes an actor in an environment of the autonomous vehicle, the actor having an associated actor type and an actor motion behavior characteristic. The augmented data may be varied to create different sets of augmented data. The sets of augmented data can be used to create one or more simulation scenarios that in turn are used to produce machine learning models to control the operation of autonomous vehicles.
1. An autonomous vehicle apparatus that comprises one or more computers to operate an autonomous vehicle using a machine learning model, wherein the machine learning model is trained at least in part using a simulation scenario, wherein the simulation scenario is generated by:
receiving logged data of the autonomous vehicle;
generating augmented data from the logged data, the augmented data describing an actor in an environment of the autonomous vehicle, the actor having an associated actor type and an actor motion behavior characteristic; and
generating the simulation scenario from the augmented data.
2. The autonomous vehicle apparatus of claim 1 , wherein the simulation scenario is also generated by:
executing a simulation based on the simulation scenario to generate a simulated output;
providing 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 autonomous vehicle apparatus of claim 1 , wherein the logged data includes raw sensor data and one of data from a video game and data from film.
4. The autonomous vehicle apparatus of claim 1 , wherein the logged data is timeseries logged data including localization data and tracking data.
5. The autonomous vehicle apparatus of claim 1 , wherein the logged data includes one of raw sensor data from any one or more sensors, state or localization data from a localization subsystem, state or perception data from a perception subsystem, state or planning data from a planning subsystem and state or control data from a control subsystem.
6. The autonomous vehicle apparatus of claim 1 , wherein the simulation scenario is also generated by:
mapping the logged data to a coordinate system to produce mapped logged data; and
performing smoothing of the mapped logged data to produce smoothed data; and
wherein the smoothed data is used in generating augmented data from the logged data.
7. The autonomous vehicle apparatus of claim 1 , wherein the augmented data comprises a simulation scenario that describes motion behavior of a simulated autonomous vehicle and at least one simulated actor.
8. The autonomous vehicle apparatus of claim 1 , wherein generating the augmented data comprises:
identifying, from the logged data, actors and generating actor states to create an initial augmented data;
sampling the initial augmented data to generate sampled augmented data; and
generating a variation of the sampled augmented data.
9. The autonomous vehicle apparatus of claim 8 , wherein the generating the variation includes changing one from a group of actor velocity, actor type, actor size, actor path, lateral offset of motion, longitudinal offset of motion, adding an actor, deleting an actor and actor behavior response.
10. The autonomous vehicle apparatus of claim 8 , wherein the generating the variation includes generating a plurality of sets of sampled augmented data, and wherein the generating the simulation scenario from the augmented data includes generating a plurality of simulation scenarios each corresponding to one set of the sets of sampled augmented data.
11. A computer-implemented method, comprising:
receiving logged data of an autonomous vehicle;
generating augmented data from the logged data, the augmented data describing an actor in an environment of the autonomous vehicle, the actor having an associated actor type and an actor motion behavior characteristic;
generating a simulation scenario from the augmented data;
executing a simulation based on the simulation scenario to produce simulation data;
training a machine learning model using the simulation data to generate a trained machine learning model; and
controlling an autonomous vehicle using the trained machine learning model.
12. The computer-implemented method of claim 11 , wherein the simulation data includes a simulated output of the simulation scenario, and training the machine learning model comprises:
providing 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 computer-implemented method of claim 11 , wherein the logged data includes raw sensor data and one of data from a video game and data from film.
14. The computer-implemented method of claim 11 , wherein the logged data is timeseries logged data including localization data and tracking data.
15. The computer-implemented method of claim 11 , wherein the logged data includes one of raw sensor data from any one or more sensors, state or localization data from a localization subsystem, state or perception data from perception subsystem, state or planning data from a planning subsystem and state or control data from a control subsystem.
16. The computer-implemented method of claim 11 , further comprising:
mapping the logged data to a coordinate system to produce mapped logged data; and
performing smoothing of the mapped logged data to produce smoothed data; and
wherein the smoothed data is used in generating augmented data from the logged data.
17. The computer-implemented method of claim 11 , wherein the simulation data comprises a simulation scenario that describes motion behavior of a simulated autonomous vehicle and at least one simulated actor.
18. The computer-implemented method of claim 11 , wherein generating the augmented data comprises:
identifying, from the logged data, actors and generating actor states to create an initial augmented data;
sampling the initial augmented data to generate sampled augmented data; and
generating a variation of the sampled augmented data.
19. The computer-implemented method of claim 18 , wherein the generating the variation includes changing one from a group of actor velocity, actor type, actor size, actor path, lateral offset of motion, longitudinal offset of motion, adding an actor, deleting an actor and actor behavior response.
20. The computer-implemented method of claim 18 , wherein the generating the variation includes generating a plurality of sets of sampled augmented data, and wherein the step of generating the simulation scenario from the augmented data includes generating a plurality of simulation scenarios each corresponding to one set of the sets of sampled augmented data.
21. A system, comprising:
a processor and a memory having computer program instructions for training a machine learning model of an autonomous vehicle, the machine learning model being trained at least in part using simulation data that generated by:
receiving logged data of the autonomous vehicle;
generating augmented data from the logged data, the augmented data describing an actor in an environment of the autonomous vehicle, the actor having an associated actor type and an actor motion behavior characteristic;
generating a simulation scenario as the simulation data, the simulation scenario generated from the augmented data; and
executing a simulation based on a simulation scenario.