Safety metric prediction
Techniques for predicting safety metrics associated with near-miss conditions for a vehicle, such as an autonomous vehicle, are discussed herein. For instance, a training system identifies an object in an environment and determines a trajectory for the object. The training system may receive a trajectory for a vehicle and associate the trajectory for the object and the trajectory for the vehicle with an event involving the object and the vehicle. In examples, the training system determines a parameter associated with motion of the vehicle as indicated by the trajectory of the vehicle relative to the trajectory of the object, and the event. Then, the training system may determine a safety metric associated with the event that indicates whether the vehicle came within a threshold of a collision with the object during a time period associated with the event.
1. A system comprising:
one or more processors; and
one or more non-transitory computer-readable media storing instructions that, when executed, cause the one or more processors to perform operations comprising:
receiving sensor data from a sensor associated with an autonomous vehicle operating in an environment;
identifying, based at least in part on the sensor data, an object in the environment;
determining, based at least in part on the sensor data, a trajectory for the object;
determining, based at least in part on the sensor data and a destination for the autonomous vehicle, a trajectory for the autonomous vehicle;
determining, based at least in part on the trajectory of the object and the trajectory of the autonomous vehicle, a possibility of a collision between the autonomous vehicle and the object;
determining a value for a parameter associated with the possibility of the collision between the autonomous vehicle and the object;
inputting the value for the parameter into a machine-learned model;
receiving, from the machine-learned model, a safety metric of an event, the event corresponding to the possibility of the collision between the autonomous vehicle and the object;
storing at least a portion of the sensor data associated with the event;
labeling the at least a portion of the sensor data based on the safety metric;
determining that the event is unassociated with a user labeled event indicating that the event is safety related; and
storing the event, with the parameter, as an unlabeled safety related event.
2. The system of claim 1 , the operations further comprising:
determining a future time to the possibility of the collision between the autonomous vehicle and the object,
wherein the parameter is based at least in part on the future time.
3. The system of claim 2 , wherein the sensor data is received from log data which comprises previously generated sensor data, and
wherein determining the value of the parameter comprises:
receiving a portion of the log data associated with a time that the sensor data was received; and
determining, from the log data, the parameter associated with the event.
4. The system of claim 1 , the operations further comprising:
determining that the autonomous vehicle disengaged from autonomous driving during a time period that includes the event; and
storing, as a disengaged event, the event with the parameter in log data.
5. The system of claim 1 , wherein the environment is a simulated environment generated for a simulation and the sensor data is simulated sensor data, the operations further comprising:
determining, based at least in part on the simulated sensor data, a first number of multiple events during the simulation associated with the parameter, the multiple events including the event;
determining a second number of multiple events associated with the parameter and the autonomous vehicle or a human driver in a real-world environment;
determining a difference between the first number and the second number; and
modifying a value associated with the parameter based at least in part on the difference.
6. A method comprising:
receiving sensor data from a sensor associated with an autonomous vehicle operating in an environment;
identifying, based at least in part on the sensor data, an object in the environment;
determining, based at least in part on the sensor data, a trajectory for the object;
determining a trajectory for the autonomous vehicle;
determining, based at least in part on the trajectory of the object and the trajectory of the autonomous vehicle, a parameter associated with a safety related event;
determining, based on the parameter, a safety metric associated with the safety related event;
determining that the safety related event is unassociated with a user labeled event indicating that the safety related event is safety related; and
storing the safety related event, with the parameter, as an unlabeled safety related event.
7. The method of claim 6 , wherein the parameter is determinable based at least in part on a possibility of a future collision and corresponds to:
an estimated time to a predicted collision between the autonomous vehicle and the object,
a deceleration required to avoid the predicted collision between the autonomous vehicle and the object, or
a proportion of stopping distance between a remaining distance from the autonomous vehicle to the object to a location of the predicted collision and a minimum acceptable stopping distance.
8. The method of claim 6 , wherein determining the safety metric comprises:
inputting the parameter into a machine-learned model comprising a gradient boosting classifier; and
receiving, from the machine-learned model, the safety metric.
9. The method of claim 6 , wherein the safety metric indicates that a collision between the autonomous vehicle and the object fails to occur in a time period associated with the safety related event.
10. The method of claim 6 , wherein the parameter corresponds to:
a lateral distance between the autonomous vehicle and the object,
a speed of the autonomous vehicle,
an estimated longitudinal distance associated with an unexpected acceleration or deceleration by the object, or
an estimated lateral distance associated with an unexpected maneuver by the object.
11. The method of claim 6 , further comprising:
determining that the autonomous vehicle disengaged from autonomous driving during a time period that includes the safety related event; and
storing, as a disengaged event, the safety related event with the parameter in log data.
12. The method of claim 6 , wherein the environment is a simulated environment generated for a simulation, the method further comprising:
determining a first number of multiple events during the simulation associated with the parameter, the multiple events including the event;
determining a second number of multiple events associated with the parameter and the autonomous vehicle or a human driver in a real-world environment;
determining a difference between the first number and the second number; and
modifying a value associated with the parameter based at least in part on the difference.
13. The method of claim 6 , further comprising:
determining that the safety related event is unassociated with the user labeled event due to one of user error or improper classification.
14. One or more non-transitory computer-readable media storing instructions that, when executed by one or more processors, perform operations comprising:
receiving sensor data from a sensor associated with an autonomous vehicle operating in an environment;
identifying, based at least in part on the sensor data, an object in the environment;
determining, based at least in part on the sensor data, a trajectory for the object;
determining a trajectory for the autonomous vehicle;
determining, based at least in part on the trajectory of the object and the trajectory of the autonomous vehicle, a parameter associated with a safety related event;
determining, based on the parameter, a safety metric associated with the safety related event;
determining that the safety related event is unassociated with a user labeled event indicating that the safety related event is safety related; and
storing the safety related event, with the parameter, as an unlabeled safety related event.
15. The one or more non-transitory computer-readable media of claim 14 , the operations further comprising:
determining a future time to a possibility of a collision between the autonomous vehicle and the object,
wherein the parameter is based at least in part on the future time.
16. The one or more non-transitory computer-readable media of claim 15 , wherein the sensor data is received from log data which comprises previously generated sensor data, and
wherein determining the parameter comprises:
receiving a portion of the log data associated with a time that the sensor data was received; and
determining, from the log data, the parameter associated with the safety related event.
17. The one or more non-transitory computer-readable media of claim 14 , wherein the parameter is determinable based at least in part on a possibility of a future collision and corresponds to:
an estimated time to a predicted collision between the autonomous vehicle and the object,
a deceleration required to avoid the predicted collision between the autonomous vehicle and the object, or
a proportion of stopping distance between a remaining distance from the autonomous vehicle to the object to a location of the predicted collision and a minimum acceptable stopping distance.
18. The one or more non-transitory computer-readable media of claim 14 , wherein determining the safety metric comprises:
inputting the parameter into a machine-learned model comprising an gradient boosting classifier; and
receiving, from the machine-learned model, the safety metric.
19. The one or more non-transitory computer-readable media of claim 14 , wherein the safety metric indicates that a collision between the autonomous vehicle and the object fails to occur in a time period associated with the safety related event.
20. The one or more non-transitory computer-readable media of claim 14 , the operations further comprising:
determining that the safety related event is unassociated with the user labeled event due to one of user error or improper classification.