Automatic mode selection system

1 . A computer implemented method for automatically selecting a mode for a vehicle, comprising:

receiving, through a video camera of the vehicle, real-time perception data comprising a sequence of image frames representing an environment in a direction of travel of the vehicle;

predicting, using a first fitted inference model, based on the sequence of image frames, a sequence of first-subsystem candidate mode predictions for a first-subsystem of the vehicle, wherein the first-subsystem candidate mode predictions correspond to a set of predefined modes for the first-subsystem, each of the predefined modes defining a respective set of one or more operating parameters for the first-subsystem;

determining, based on the sequence of first-subsystem candidate mode predictions, a first-subsystem mode; and

causing the first-subsystem mode to be implemented by the first-subsystem.

2 . The method of claim 1 wherein determining the first-subsystem mode comprises:

filtering the sequence of first-subsystem candidate mode predictions to identify a possible first-subsystem mode prediction; and

selecting the first-subsystem mode based on the possible first-subsystem mode prediction.

3 . The method of claim 2 wherein each of the first-subsystem candidate mode predictions comprises a candidate mode probabilities vector that includes a respective probability value for each of the predefined modes for the first-subsystem,

wherein filtering the predicted sequence of first-subsystem candidate mode predictions comprises applying an impulse response filter to the sequence of first-subsystem candidate mode predictions, an output of the impulse response filter comprising the possible first-subsystem mode prediction.

4 . The method of claim 3 wherein the filtering is based on a most recent prediction in the sequence of first-subsystem candidate mode predictions, past predictions in the sequence of first-subsystem candidate mode predictions and past possible first-subsystem mode predictions output by the filtering.

5 . The method of claim 3 wherein filtering applies low pass filtering to the sequence of first-subsystem candidate mode predictions.

6 . The method of claim 3 wherein determining the first-subsystem mode further comprises performing a time check to confirm that a time threshold has passed since a last first-subsystem mode was caused to be implemented, and when the time threshold has not passed, causing the receiving, predicting and determining to be repeated without causing the first-subsystem mode to be implemented by the first-subsystem.

7 . The method of claim 3 wherein selecting the first-subsystem mode based on the possible first-subsystem mode prediction comprises performing a hysteresis decision based on a most recent first-subsystem mode that was caused to be implemented.

8 . The method of claim 7 wherein performing the hysteresis decision comprises:

computing, for each of the predefined modes, a mode-specific difference between the probability value for each of the predefined modes in the first-subsystem mode prediction and a probability value determined for the most recent first-subsystem mode that was caused to be implemented;

computing for each mode-specific difference that has a value greater than a predefined mode-specific hysteresis value that corresponds to the most recent first-subsystem mode that was caused to be implemented, a difference between the mode-specific difference and the predefined mode-specific hysteresis value; and

selecting, as the first-subsystem mode, a mode associated with a highest computed difference between the mode-specific difference and the predefined mode-specific hysteresis value.

9 . The method of claim 3 , further comprising pre-processing the real-time perception data prior to the predicting, the pre-processing comprising: (i) extracting a subset of image frames from the sequence of image frames; and (ii) resizing and cropping image frames included in the extracted subset of image frames, wherein the predicting comprises predicting a respective first-subsystem candidate mode prediction for each of the resized and cropped image frames included in the extracted subset of image frames.

10 . The method of claim 9 further comprising receiving, through a vehicle sensor system, real-time vehicle state data indicating current operating characteristics of the vehicle, wherein at least one of a filtering rate used for the filtering and a rate used for extracting the subset of image frames, is based on a vehicle speed included in the real-time vehicle state data.

11 . The method of claim 1 wherein the first-subsystem corresponds to a driveline system of the vehicle and the set of predefined modes comprises a four wheel drive mode and a two wheel drive mode.

12 . The method of claim 1 wherein the first-subsystem corresponds to a suspension system of the vehicle and the set of predefined modes comprises a first mode, a second mode and a third mode, where the first mode corresponds to a softer ride than the second mode and the second mode corresponds to a softer ride than the third mode.

13 . The method of claim 1 further comprising pre-training the first fitted inference model, the pre-training comprising:

obtaining a time-stamped sequence of image frames captured by a forward-facing camera video camera of a test vehicle during operation of the test vehicle by an expert driver;

obtaining a time-stamped sequence of mode inputs performed by the expert driver contemporaneously with the capturing of the image frames;

merging the sequence of the images frames and the mode inputs to provide a set of labelled training image frames; and

training a convolution neural network based on the labelled training image frames to obtain the first fitted inference model.

14 . The method of claim 1 wherein determining the first-subsystem mode comprises:

identifying, based on the sequence of first-subsystem candidate mode predictions, a possible first-subsystem mode that is appropriate for real-time implementation by the first-subsystem; and

confirming that the possible first-subsystem can be applied as the first-subsystem mode based on a time duration since a most recently implemented first-subsystem candidate mode.

15 . The method of claim 14 wherein identifying a possible first-subsystem candidate mode that is appropriate comprises confirming that the real-time implementation will meet specified vehicle criteria, wherein the vehicle criteria specifies one or more operating constraints at which the first-subsystem candidate mode can be implemented, the operating constraints including one or more of a maximum speed, operating state of one or more other vehicle subsystems, and a direction of travel.

16 . A computer implemented method for automatically selecting a mode for a vehicle, comprising:

receiving, through a perception system of the vehicle, real-time perception data representing an environment in a desired direction of travel of the vehicle;

predicting, using a first fitted inference model, based on the real-time perception data, a sequence of first-subsystem candidate mode predictions for a first-subsystem of the vehicle, wherein each of the first-subsystem candidate mode predictions corresponds to a set of predefined modes for the first-subsystem, each of the predefined modes defining a respective set of one or more operating parameters for the first-subsystem;

determining, based on the sequence of first-subsystem candidate mode predictions, a first-subsystem mode; and

causing the first-subsystem mode to be implemented by the first-subsystem.

17 . A vehicle equipped with a computer system, the computer system being configured for automatically selecting an operating mode for the vehicle by:

receiving, through a video camera of the vehicle, a sequence of image frames representing upcoming terrain in a direction of travel of the vehicle;

predicting, using a first fitted inference model, based on the sequence of image frames, a sequence of candidate mode predictions for the vehicle, wherein the candidate mode predictions correspond to a set of predefined modes for a system of the vehicle, each of the predefined modes defining a respective set of one or more operating parameters for the system;

determining, based on the sequence of candidate mode predictions, a system mode; and

causing the system mode to be implemented by the vehicle.

18 . The vehicle of claim 17 , wherein determining the system mode comprises:

filtering the sequence of candidate mode predictions to identify a possible system mode prediction; and

selecting the system mode based on the possible system mode prediction.

19 . The vehicle of claim 18 wherein each of the candidate mode predictions comprises a candidate mode probabilities vector that includes a respective probability value for each of the predefined modes,

wherein filtering the sequence of candidate mode predictions comprises applying an impulse response filter to the sequence of candidate mode predictions, an output of the impulse response filter comprising the possible system mode prediction.

20 . The vehicle of claim 19 wherein the filtering is based on a most recent prediction in the sequence of candidate mode predictions, past predictions in the sequence of candidate mode predictions and past possible system mode predictions output by the filtering.