Hardware fault detection for feedback control systems in autonomous machine applications
Systems and methods for detecting hardware faults in computer-based feedback control systems. Multiple instances of the system control program(s) are run on system processors. System sensor data are input to each instance, and the control commands output by each instance are compared. As instantiations of the same programs receive largely the same sensor data, differences between output commands may indicate the presence of one or more hardware faults.
1. A method comprising:
receiving sensor data from at least one sensor of a system having at least one actuator;
transmitting a first input set and a second input set each corresponding to the sensor data to at least two instances of a control program for the system, each instance of the at least two instances comprising a different process being executed by processing circuitry of the system on a same system on a chip,
a first instance of the at least two instances being configured to receive the first input set corresponding to a first subset of the sensor data to generate first output indicating a first control command for the system, and
a second instance of the at least two instances being configured to receive the second input set corresponding to a second subset of the sensor data that is different from the first subset to generate second output indicating a second control command; and
comparing the first output from the first instance to the second output from the second instance to determine whether a hardware fault is present in the processing circuitry.
2. The method of claim 1 , wherein the transmitting includes transmitting successive units of the sensor data to alternating ones of the at least two instances such that the first instance processes the first subset of the sensor data using a first processing frequency and the second instance processes the second subset of the sensor data using a second processing frequency that is offset relative to the first processing frequency.
3. The method of claim 1 , wherein the first instance processes a first stream of input data corresponding to the first subset of the sensor data to generate the first output and the second instance processes a second stream of input data corresponding to the second subset of the sensor data to generate the second output.
4. The method of claim 1 , wherein when the comparing indicates that the first control command is a same control command as the second control command, the hardware fault is determined to not be present.
5. The method of claim 1 , wherein the at least two instances comprise three or more instances, and wherein the comparing includes determining whether the hardware fault is present according to outputs of a majority of the three or more of the instances.
6. The method of claim 1 , wherein the comparing includes determining whether the hardware fault is present when corresponding outputs of the at least two instances differ from each other by greater than a predetermined amount.
7. The method of claim 1 , wherein the comparing includes determining whether the hardware fault is present according to one or more machine learning models having input corresponding to the first output and the second output and output indicating a determination of the hardware fault.
8. The method of claim 1 , wherein the comparing includes determining whether the hardware fault is present according to first multiple outputs generated by the first instance and second multiple outputs generated by the second instance.
9. The method of claim 1 , wherein the at least two instances of the control program are executed by different virtual devices.
10. The method of claim 1 , wherein the system comprises an autonomous vehicle.
11. The method of claim 1 , wherein the processing circuitry comprises parallel processing circuitry.
12. A system comprising:
at least one processor comprising processing circuitry to:
execute at least two instances of a control program, the control program for generating, from sensor output obtained using at least one sensor, output indicating control commands for controlling a machine; and
compare the output from the at least two instances of the control program to determine whether a hardware fault is present in the processing circuitry according to one or more machine learning models having input corresponding to the output from the at least two instances and output indicating a determination of the hardware fault.
13. The system of claim 12 , wherein successive units of the sensor output are provided to alternating ones of the at least two instances to generate the output from the at least two instances.
14. The system of claim 12 , wherein alternating units of the sensor output are provided to each of the at least two instances to generate the output from the at least two instances.
15. The system of claim 12 , wherein the at least two instances comprise three or more of the instances, and wherein the comparing includes determining whether the hardware fault is present according to outputs of a majority of the three or more of the instances.
16. The system of claim 12 , wherein the comparing includes determining whether the hardware fault is present when corresponding outputs of the at least two instances different from each other by greater than a predetermined amount.
17. The system of claim 12 , wherein the comparing includes determining whether the hardware fault is present according to multiple outputs of each of the at least two instances.
18. A processor comprising:
one or more circuits to:
transmit successive units from a stream of sensor data obtained using one or more sensors of a machine to alternating ones of at least two instances of a control program for the machine to generate a first stream of input data and a second stream of input data, the at least two instances being executed using processing circuitry of the machine,
a first instance of the at least two instances being configured to process the first stream of input data corresponding to the sensor data using a first processing frequency to generate first output indicating one or more first system control commands for the machine, and
a second instance of the at least two instances being configured to process the second stream of input data corresponding to the sensor data using a second processing frequency that alternates with respect to the first processing frequency to generate second output indicating one or more second system control commands for the machine; and
compare the first output from the first instance to the second output from the second instance to determine whether a hardware fault is present in the processing circuitry.
19. The processor of claim 18 , wherein when the comparing indicates that the one or more first system control commands includes same control commands as the one or more second system control commands, the hardware fault is determined to not be present.
20. The processor of claim 18 , wherein one or more machine learning models having input corresponding to the first output and the second output and output indicating a determination of the hardware fault are used to determine whether the hardware fault is present.
21. The processor of claim 18 , wherein the first instance and the second instance are to process different subsets of the stream of sensor data to generate the first output and the second output.
22. The processor of claim 18 , wherein the first instance and the second instance are executed on a single system on a chip.