Detection of hazardous driving using machine learning
An autonomous driving system could create or exacerbate a hazardous driving situation due to incorrect machine learning, algorithm design, sensor limitations, environmental conditions or other factors. This technology presents solutions that use machine learning to detect when the autonomous driving system is in this state e.g., erratic or reckless driving and other behavior, in order to take remedial action to prevent a hazard such as a collision.
1 . A processing system comprising:
one or more sensors that provide one or more sensor outputs for detecting one or more objects in one or more sensory fields corresponding to the one or more sensors; and
at least one processor operatively coupled to the one or more sensors, the at least one processor to execute:
a first artificial intelligence object to generate one or more plans based at least on processing sensor data obtained using the one or more sensors;
a second artificial intelligence object to generate, at least partially in parallel with the first artificial intelligence object generating the one or more plans, at least one of a hazard level or a safety level for individual plans of the one or more plans, the second artificial intelligence object processing at least a subset of the sensor data obtained using the one or more sensors to generate the at least one of the hazard level or the safety level; and
an arbiter to:
receive at least one of the hazard levels or the safety levels corresponding to the one or more plans and one or more control signals corresponding to the one or more plans; and
select at least one control signal of the one or more control signals based at least on at least one of the hazard levels or the safety levels corresponding to the one or more plans; and
one or more control components to cause navigation of an autonomous or semi-autonomous machine based at least on the selected at least one control signal,
wherein the use of the first artificial intelligence object, the second artificial intelligence object, and the arbiter decomposes a safety requirement to achieve automotive safety integrity level (ASIL) D.
2 . The processing system of claim 1 wherein the executing second artificial intelligence object ranks the one or more plans based on hazard level or safety level, and the arbiter is preprogrammed to selectively select and use ranked plans based on the hazard levels or the safety levels.
3 . The processing system of claim 1 wherein the executing second artificial intelligence object generates, based at least on the hazard levels or safety levels corresponding to the one or more plans, one or more hazard signals and applies the one or more hazard signals to the first artificial intelligence object, the executing first artificial intelligence object configured to adapt at least one individual plan in response to the one or more hazard signals such that at least one of a lower hazard level or a higher safety level is generated by the second artificial intelligence object to correspond to the adapted at least one individual plan.
4 . The processing system of claim 3 wherein the arbiter rejects plans the first artificial intelligence object provides when the first artificial intelligence object fails to adapt the at least one individual plan in response to the hazard signals such that at least one of a lower hazard level or a higher safety level is generated by the second artificial intelligence object to correspond to the adapted at least one individual plan.
5 . The processing system of claim 1 wherein the second artificial intelligence object provides short-term path safety evaluation.
6 . The processing system of claim 5 wherein the second artificial intelligence object provides short-term path correction based on the short-term path safety evaluation it performs.
7 . The processing system of claim 1 wherein the arbiter temporarily substitutes safe control outputs from the second artificial intelligence object for hazardous control outputs from the first artificial intelligence object.
8 . The processing system of claim 1 wherein the second artificial intelligence object alerts a human if it detects the first artificial intelligence object generates unsafe control outputs.
9 . The processing system of claim 1 wherein the sensor outputs comprise at least two of object location/position tracking, environment sensing, obstacle location mapping, or driving mode.
10 . The processing system of claim 1 wherein the second artificial intelligence object operates at a lower resolution than the first artificial intelligence object.
11 . A processing method performed by an apparatus including one or more sensors that provide one or more sensor outputs for detecting one or more objects, and at least one processor operatively coupled to the sensors, the at least one processor executing stored program instructions to perform operations comprising:
executing a first artificial intelligence object to generate one or more plans based at least on processing sensor data obtained using the one or more sensors,
executing a second artificial intelligence object to generate, at least partially in parallel with the first artificial intelligence object generating the one or more plans, at least one of a hazard level or a safety level for individual plans of the one or more plans, the second artificial intelligence object processing at least a subset of the sensor data obtained using the one or more sensors to generate the at least one of the hazard level or the safety level,
receiving, with an arbiter, at least one of the hazard levels or the safety levels corresponding to the one or more plans and one or more control signals corresponding to the one or more plans,
selecting, with the arbiter, at least one control signal based at least on at least one of the hazard levels or the safety levels corresponding to the one or more plans, and
operating one or more control components to cause navigation of an autonomous or semi-autonomous machine based at least on the selected at least one control signal,
wherein the use of the first artificial intelligence object, the second artificial intelligence object, and the arbiter decomposes a safety requirement to achieve automotive safety integrity level (ASIL) D.
12 . The processing method of claim 11 further including:
ranking the one or more plans based on respective the hazard level or the safety level corresponding to the one or more plans, and
selectively selecting and using ranked individual prioritized plans of the one or more plans based on the respective the hazard level or the safety level.
13 . The processing method of claim 11 further including generating hazard signals and applying the hazard signals to the first artificial intelligence object, the first artificial intelligence object adapting individual plans in response to the hazard signals to cause the second artificial intelligence object to generate at least one of lower hazard levels or higher safety levels with respect to the adapted individual plans.
14 . The processing method of claim 13 further including rejecting individual plans the first artificial intelligence object provides if the first artificial intelligence object fails to adapt the individual plans in response to the hazard signals to cause the second artificial intelligence object to generate at least one of lower hazard levels or higher safety levels with respect to the adapted individual plans.
15 . The processing method of claim 11 further including the second artificial intelligence object providing short-term path safety evaluation.
16 . The processing method of claim 15 further including the second artificial intelligence object providing short-term path correction based on the short-term path safety evaluation it performs.
17 . The processing method of claim 11 further including temporarily substituting safe control outputs from the second artificial intelligence object for unsafe control outputs from the first artificial intelligence object.
18 . The processing method of claim 11 further including the second artificial intelligence object alerting a human if it detects the first artificial intelligence object generates unsafe control outputs.
19 . The processing method of claim 11 wherein the sensor outputs comprise at least two of object location/position tracking, environment sensing, obstacle location mapping, or driving mode.
20 . The processing method of claim 11 further including operating the second artificial intelligence object at a lower resolution than the first artificial intelligence object.
21 . The processing system of claim 1 wherein the at least one processor comprises
a first processor configured to execute the first artificial intelligence object,
a second processor configured to execute the second artificial intelligence object, and
a third processor configured to select the at least one control signal in accordance with the hazard levels or the safety levels, and to apply the selected at least one control signal to an actuator of the machine.