Generating perception scenarios for an autonomous vehicle from simulation data
Simulation data of the autonomous vehicle is processed by executing a simulation based on the simulation data to generate a simulation result. Then a perception scenario is generated from the simulation result. The generated perception scenario is validated by verifying whether a constraint is satisfied to produce a validated perception scenario. The validated perception scenario can be used to create or refine a perception model used for controlling the operation of autonomous vehicles.
1. A method for an autonomous vehicle, the method comprising:
receiving simulation data including the autonomous vehicle, the simulation data generated based on random sampling of snippets of logged data from a set of snippets of logged data associated with real-world driving, the random sampling of snippets of logged data based on an identifier identifying a specific characteristic in a snippet of logged data;
executing a first simulation of a planning subsystem of the autonomous vehicle based on the simulation data to generate a simulation result;
executing a second simulation of a perception subsystem of the autonomous vehicle using the simulation result as an input to generate an amended simulation result including ground truth data based on dynamic state information of one or more actors in the second simulation;
generating a perception scenario using the amended simulation result; and
validating the perception scenario by verifying whether a constraint is satisfied to produce a validated perception scenario.
2. The method of claim 1 , further comprising:
providing the validated perception scenario as a training input to machine learning engine to generate a perception model and a predicted output of the perception model; and
updating one or more weights in the perception model based on a difference between the predicted output and the amended simulation result.
3. The method of claim 2 , wherein the perception scenario is for a lidar sensor of the autonomous vehicle, and the perception model is for the lidar sensor.
4. The method of claim 2 , wherein the perception scenario is for a radar sensor of the autonomous vehicle and the perception model is for the radar sensor.
5. The method of claim 2 , wherein the perception scenario is for a camera of the autonomous vehicle and the perception model is for the camera.
6. The method of claim 1 , wherein the perception scenario is for a plurality of sensors of the autonomous vehicle, and the plurality of sensors are from a group of a lidar sensor, a radar sensor and a camera.
7. The method of claim 1 , wherein the validating the perception scenario verifies a tracking constraint.
8. The method of claim 1 , wherein the simulation data is generated from logged sensor data.
9. The method of claim 8 , wherein the logged sensor data is from a plurality of sensors.
10. The method of claim 1 , wherein the simulation data is generated from data from a simulation or a video game.
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 the execution of the instructions by one or more processors, cause the one or more processors to perform the following operations:
receiving simulation data including an autonomous vehicle, the simulation data generated based on random sampling of snippets of logged data from a set of snippets of logged data associated with real-world driving, the random sampling of snippets of logged data based on an identifier identifying a specific characteristic in a snippet of logged data;
executing a first simulation of a planning subsystem of the autonomous vehicle based on the simulation data to generate a simulation result;
executing a second simulation of a perception subsystem of the autonomous vehicle using the simulation result as an input to generate an amended simulation result including ground truth data based on dynamic state information of one or more actors in the second simulation;
generating a perception scenario using the amended simulation result; and
validating the perception scenario by verifying whether a constraint is satisfied to produce a validated perception scenario.
12. The system of claim 11 , wherein the operations further comprise:
providing the validated perception scenario as a training input to machine learning engine to generate a perception model and a predicted output of the perception model; and
updating one or more weights in the perception model based on a difference between the predicted output and the amended simulation result.
13. The system of claim 12 , wherein the perception scenario is for a lidar sensor of the autonomous vehicle, and the perception model is for the lidar sensor.
14. The system of claim 12 , wherein the perception scenario is for a radar sensor of the autonomous vehicle and the perception model is for the radar sensor.
15. The system of claim 12 , wherein the perception scenario is for a camera of the autonomous vehicle and the perception model is for the camera.
16. The system of claim 11 , wherein the perception scenario is for a plurality of sensors of the autonomous vehicle, and the plurality of sensors are from a group of a lidar sensor, a radar sensor and a camera.
17. The system of claim 11 , wherein the validating the perception scenario verifies a tracking constraint.
18. The system of claim 11 , wherein the simulation data is generated from logged sensor data and the logged sensor data is from a plurality of sensors.
19. The system of claim 11 , wherein the simulation data is generated from data from a simulation or a video game.
20. A non-transitory computer readable storage medium storing computer instructions executable by one or more processors to perform a method of generating a perception model for an autonomous vehicle, the method comprising:
receiving simulation data including the autonomous vehicle, the simulation data generated based on random sampling of snippets of logged data from a set of snippets of logged data associated with real-world driving, the random sampling of snippets of logged data based on an identifier identifying a specific characteristic in a snippet of logged data;
executing a first simulation of a planning subsystem of the autonomous vehicle based on the simulation data to generate a simulation result;
executing a second simulation of a perception subsystem of the autonomous vehicle using the simulation result as an input to generate an amended simulation result including ground truth data based on dynamic state information of one or more actors in the second simulation;
generating a perception scenario using the amended simulation result; and
validating the perception scenario by verifying whether a constraint is satisfied to produce a validated perception scenario.