Collision avoidance/mitigation by machine learning and automatic intervention
A first vehicle in traffic can use machine learning and artificial intelligence to detect an imminent collision with a second vehicle or other object. A well-trained AI algorithm can select a sequence of actions (braking, swerving, or accelerating—depending on the specific kinetics) to avoid the collision if possible, and to reduce or minimize the harm if unavoidable. With proper training, the AI model may also infer the intent and future actions of the second vehicle, as well as potential interference of other traffic agents. A good algorithm can also infer the intent of the driver of the first vehicle, for example based on prior driving habits. The AI algorithm may be implemented in a processor on the subject vehicle, potentially in communication with another processor at a fixed site such as a local access point or a central supercomputer. With super-fast AI solutions, lives will be saved!
1. A non-transitory computer readable medium, comprising instructions for causing a computing environment to perform a method for mitigating vehicle collisions, the non-transitory computer readable medium comprising an algorithm employing machine learning or artificial intelligence, the algorithm based on prior learning, the method comprising:
detecting, with a sensor mounted on a subject vehicle, a second vehicle, and calculating, by the algorithm employing machine learning or artificial intelligence, whether the second vehicle will collide with the subject vehicle;
calculating, by the algorithm employing machine learning or artificial intelligence, whether a first set of actions would avoid a collision with the subject vehicle;
if the first set of actions would avoid the collision, implementing the first set of actions by actuating linkages that control steering, acceleration, or braking of the subject vehicle according to the first set of actions;
if the first set of actions would fail to avoid the collision, calculating, by the algorithm employing machine learning or artificial intelligence, whether a second set of actions would avoid the collision:
if the second set of actions would avoid the collision, implementing the second set of actions by actuating the linkages that control steering, acceleration, or braking of the subject vehicle according to the second set of actions; and
if the second set of actions would fail to avoid the collision, calculating, by the algorithm employing machine learning or artificial intelligence, which of the first or second set of actions, or a third set of actions, will cause a least amount of harm, and implementing which of the first or second set of actions, or the third set of actions, that causes the least amount of harm by actuating the linkages that control steering, acceleration, or braking of the subject vehicle according to the first, second, or third set of actions;
wherein the first or second sequences set of actions are further determined at least in part based on the algorithm employing machine learning or artificial intelligence, the algorithm based on prior learning.
2. The non-transitory computer-readable medium of claim 1 , wherein the method is such that the least amount of harm is a local minimum or an absolute minimum of harm.
3. The non-transitory computer-readable medium of claim 1 , wherein the method is such that the least amount of harm is a smallest amount of harm among sets of actions yet calculated.
4. The non-transitory computer-readable medium of claim 1 , wherein the method is such that the least amount of harm is a lowest amount of harm achievable by accelerating, decelerating, and/or steering the subject vehicle.