Operating a vehicle according to an artificial intelligence model
Vehicles can be operated according to an artificial intelligence model contained in an on-board processor. The AI model can analyze sensor data, such as visible or infrared images of traffic, and determine when a collision is possible, whether it has become imminent, and whether the collision is avoidable or unavoidable using sequences of accelerations, braking, and steering. The AI model can also select the most appropriate sequence of actions from a large plurality of calculated sequences to avoid the collision if avoidable, and to minimize the harm if unavoidable. The AI model can also cause a processor to actuate linkages connected to the throttle (or electric power control), brakes (or regenerative braking), and steering to implement the selected sequence of actions. Thus the collision can be avoided or mitigated by an ADAS system or a fully autonomous vehicle.
1. Non-transitory computer-readable media in a first vehicle, the media containing an artificial intelligence (AI) model and instructions that, when executed by a computing environment, cause a method to be performed, the method comprising:
a) operating a first vehicle in an AI mode, the AI mode including:
i) using the AI model of the first vehicle to analyze surrounding objects including surrounding vehicles;
ii) using the AI model of the first vehicle to perform predictive modeling to predict future positions of the first vehicle and the surrounding objects including surrounding vehicles;
iii) using received traffic data as an input to the AI model of the first vehicle;
iv) collecting and determining data about roads from lane lines or other indicators of a roadway ahead, and using the collected and determined data as inputs to the AI model of the first vehicle;
v) wherein in the AI model of the first vehicle is used at least in part to provide an early warning of emerging hazards;
b) acquiring, using a sensor in or on the first vehicle, data about a second vehicle;
c) determining that a collision between the first and second vehicles is possible;
d) providing, as input to the AI model of the first vehicle, the data about the second vehicle;
e) determining, according to output from the AI model of the first vehicle, a sequence of actions that, when implemented by the first vehicle, is calculated to avoid the collision or reduce harm caused by the collision; and
f) wherein the AI model of the first vehicle, or a processor associated with the AI model of the first vehicle, is operably connected to an accelerator, a brake, and a steering linkage of the first vehicle, and is configured to execute the sequence of actions automatically.
2. The non-transitory computer-readable media of claim 1 , wherein the AI mode further includes using artificial intelligence to analyze driver behavior of the first vehicle to determine patterns, and using the determined patterns as an input to the AI model of the first vehicle.
3. The non-transitory computer-readable media of claim 2 , wherein the AI mode further includes using artificial intelligence to analyze driver behavior of one or more of the surrounding vehicles, or surrounding object behavior, and using the analyzed behavior as an input to the AI model of the first vehicle.
4. The non-transitory computer-readable media of claim 1 , wherein the AI mode further includes using the AI model of the first vehicle to determine, according to images from one or more visible-light cameras in or on the first vehicle, a distance between the first and second vehicles.
5. The non-transitory computer-readable media of claim 1 , wherein the AI mode further includes using the AI model of the first vehicle to operate the accelerator or the brake or the steering linkage, or combinations of these, responsive to determining, according to visible-light images, that the first vehicle is calculated to strike an object.
6. The non-transitory computer-readable media of claim 1 , wherein the AI mode further includes using artificial intelligence to determine, according to the collected and determined data, that the roadway is expected to curve, and to operate the steering linkage according to the expected curve.
7. The non-transitory computer-readable media of claim 1 , wherein the AI mode further includes using artificial intelligence to determine that the second vehicle has accelerated or has changed an acceleration, and to determine, according to further output from the AI model of the first vehicle, an intention of an operator of the second vehicle according to the acceleration or the change in acceleration.
8. The non-transitory computer-readable media of claim 1 , wherein the AI mode further includes using the AI model of the first vehicle to determine whether the possible collision between the first and second vehicles is avoidable or unavoidable, wherein the possible collision is avoidable when the possible collision can be avoided by accelerating, braking, or steering the first vehicle, and is unavoidable otherwise.
9. The non-transitory computer-readable media of claim 8 , wherein the AI mode further includes using the AI model of the first vehicle to determine, when the possible collision is avoidable, a sequence of accelerating, braking, or steering, or combinations of these, to avoid the possible collision.
10. The non-transitory computer-readable media of claim 8 , wherein the AI mode further includes using the AI model of the first vehicle to determine, when the possible collision is unavoidable, a sequence of accelerating, braking, or steering, or combinations of these, to reduce or minimize the harm caused by the possible collision.
11. The non-transitory computer-readable media of claim 1 , wherein the AI mode further includes using the AI model of the first vehicle to determine a plurality of different sequences of actions, each action comprising an accelerating action, or a braking action, or a steering action, or combinations of these, and to determine which sequence of actions of the plurality can avoid the collision or, when the collision is determined to be unavoidable, can reduce the harm caused by the collision.
12. A system, comprising:
a) a subject vehicle comprising non-transitory computer-readable media containing an artificial intelligence model;
b) one or more sensors mounted in or on the subject vehicle, each sensor configured to acquire sensor data related to a second vehicle proximate to a subject vehicle; and
c) one or more processors programmed to perform the following steps:
i) provide the sensor data as input to the artificial intelligence model;
ii) use the artificial intelligence model to determine, according to the sensor data, one or more of a position, a velocity, or an acceleration, or combinations of these, of the second vehicle;
iii) use the artificial intelligence model to determine whether a collision is imminent between the subject vehicle and the second vehicle;
iv) responsive to determining that the collision is imminent, use the artificial intelligence model to determine a plurality of sequences of actions, each action comprising an accelerating action or a braking action or a steering action or combinations of these;
v) use the artificial intelligence model to select a particular sequence of actions of the plurality, the particular sequence of actions selected to avoid the collision if avoidable and to minimize harm of the collision if unavoidable; and
vi) implement the particular sequence of actions.
13. The system of claim 12 , wherein the one or more sensors comprise one or more infrared or visible-light cameras and the sensor data comprises infrared or visible-light images of traffic proximate to the subject vehicle.
14. The system of claim 12 , wherein the steps further comprise:
a) use the artificial intelligence model to determine whether the imminent collision can be avoided by sequentially applying the accelerating, braking, and steering actions, and to determine that the imminent collision is unavoidable otherwise.
15. The system of claim 14 , wherein the steps further comprise:
a) upon determining that the imminent collision is unavoidable, use the artificial intelligence model to determine an amount of expected harm associated with each sequence of actions, and select, as the particular sequence of actions, the sequence of actions associated with a smallest amount of expected harm.
16. Non-transitory computer-readable media comprising an artificial intelligence model and instructions for causing a computing environment mounted on a first vehicle to perform a method for mitigating an imminent collision, the method comprising:
a) receiving a signal from a sensor mounted in or on the first vehicle;
b) using the artificial intelligence model to detect an imminent collision between the first vehicle and a second vehicle according to the received signal;
c) using the artificial intelligence model to determine a first sequence of actions to avoid an avoidable collision when the imminent collision is unavoidable, and using the artificial intelligence model to determine a second sequence of actions to minimize harm of an unavoidable collision when the imminent collision is unavoidable, wherein the harm of the unavoidable collision comprises an estimated number of fatalities, and an estimated number of injuries, and an estimated amount of property damage resulting from the unavoidable collision;
d) causing a throttle, brakes, or steering, or a combination of these, to change a motion of the first vehicle according to the first or second sequence of actions; and
e) wherein each sequence of actions, of the first and second sequences of actions, comprises a plurality of separate sequential predetermined accelerating, braking, or steering actions, or combinations of these, by the first vehicle.
17. The non-transitory computer-readable media of claim 16 , the method further comprising using the artificial intelligence model to analyze past behavior of a driver of the first vehicle to determine patterns, and providing the determined patterns as input to further calculations by the artificial intelligence model.
18. The non-transitory computer-readable media of claim 16 , the method further comprising using the artificial intelligence model to analyze behavior of a driver of the second vehicle, and providing the analyzed behavior as input to further calculations by the artificial intelligence model.
19. The non-transitory computer-readable media of claim 16 , the method further comprising using the artificial intelligence model to analyze behavior of a pedestrian, and providing the analyzed behavior as input to further calculations by the artificial intelligence model.
20. The non-transitory computer-readable media of claim 16 , the method further comprising using the artificial intelligence model to determine a third sequence of actions while the first or second sequence of actions is being implemented, and to determine whether the third sequence of actions can avoid the collision or can cause less harm than the first or second sequence of actions.