Abstract: Scheduling state transitions in an autonomous vehicle, including: detecting a transition signal for transitioning from a first state associated with a first machine learning model to a second state associated with a second machine learning model; determining whether a precondition for generating output by the second machine learning model has been satisfied; and delaying, in response for the precondition not being satisfied, a transition from the first state to the second state.

Abstract: Processing predictive input data in an autonomous vehicle, including: receiving input data for a model; determining whether the input data for the model comprises an indication that the input data was generated based on some amount of predicted data; generating, by the model and based on the input data, output data by modifying, in response to the input data comprising the indication, one or more thresholds or one or more confidence scores of the model used in generating output data; and causing an autonomous vehicle to perform one or more driving decisions based on the output data of the model.

Abstract: Modifying autonomous vehicle sensors cameras based on detected lighting boundaries, including: determining, based on video data, that a vehicle will pass through a lighting boundary; determining, based on the video data, that an illumination of a predicted input to a camera of the vehicle after passing the lighting boundary falls outside a range of illumination based on a current exposure setting of the camera; and modifying, in response to determining that the illumination of the predicted input falls outside the range of illumination, an exposure setting of one or more cameras.

Abstract: Disengagement of an autonomous driving mode based on vehicle operating conditions, including: receiving, from a steering torque sensor, torque sensor data indicating an amount of torque applied to a steering system of an autonomous vehicle; determining a predicted torque based on one or more motion attributes of the steering system of the autonomous vehicle; determining an estimated operator-provided torque as a differential between the predicted torque and the amount of torque; selecting a threshold from a plurality of thresholds, wherein each threshold of the plurality of thresholds corresponds to a different possible operating condition of the autonomous vehicle; and disengaging an autonomous driving mode of the autonomous vehicle based on the differential exceeding the threshold.

Abstract: Reducing camera throughput to downstream systems using an intermediary device, including: receiving, by a device and from a camera via a first data link between the device and the camera, a frame; selecting, by the device, an area of focus for the frame; generating, by the device from the frame, a downsampled frame and a cropped frame, wherein the cropped frame is based on the area of focus; and providing, by the device to a computing system of an autonomous vehicle via a second data link between the device and the computing system, the downsampled frame and the cropped frame instead of the frame, wherein the second data link has a lower bandwidth than the first data link.

Abstract: Anticipatory vehicle seat actuation, including: determining, based on sensor data capturing an environment around a vehicle, a predicted environmental state corresponding to a future time and comprising the vehicle entering a turn; determining, based on the predicted environmental state, an actuation state for one or more ECUs controlling a seat state the vehicle; and configuring the seat state of the vehicle according to the actuation state before the future time.

Abstract: Fault state recovery using predefined configurations, including: determining that a first node of a plurality of nodes has failed; identifying, based on the first node failing, a failure state from a plurality of predefined failure states each corresponding to a different combination of one or more failed nodes, wherein each of the plurality of predefined failure states corresponds to a respective configuration of a plurality of configurations; and applying the respective configuration corresponding to the identified failure state, wherein applying the respective configuration comprises at least one of: updating one or more memory mapping tables based on the respective configuration or copying data from memory locations used based on a previous configuration to memory locations indicated in the respective configuration.

Abstract: A failover system for autonomous vehicles, including: detecting an error associated with a hardware resource of a first autonomous driving system of the autonomous vehicle; performing a failover from the first autonomous driving system to a second autonomous driving system comprising fewer hardware resources relative to the first autonomous driving system, wherein performing the failover causes control operations generated by the second autonomous driving system to be applied to the autonomous vehicle instead of control operations generated by the first autonomous driving system; and reconfigure one or more other hardware resources of the first autonomous driving system not associated with the error into a redundant autonomous driving system.

Abstract: Transformations in fused multiply-add instructions, including: receiving an extended fused multiply-add (FMA) instruction comprising: a first subset of bits corresponding to each operand of a fused multiply-add (FMA) operation, and a second subset of bits comprising an opcode for the extended FMA instruction, wherein the opcode identifies extended FMA instruction from an instruction set of a plurality of extended FMA instructions each having a different predefined opcode corresponding to a different combination of transformation-operand groupings; and performing, based on the opcode of the extended FMA instruction, the one or more transformations and the FMA operation.

Abstract: Selective training of neural networks using motion estimation, including: selecting, from a plurality of pixels in video data from a vehicle, based on motion relative to the vehicle, one or more pixels; and training a neural network based on the video data by zeroing out an error function applied to the selected one or more pixels.

Abstract: Autonomous vehicle model training and validation using low-discrepancy sequences may include: generating a low-discrepancy sequence in a multidimensional space comprising a plurality of multidimensional points; mapping each sample of a plurality of samples of a data corpus to a corresponding entry in the low-discrepancy sequence, wherein each sample of the plurality of samples comprises one or more environmental descriptors for an environment relative to a vehicle and one or more state descriptors describing a state of the vehicle; selecting, from the data corpus, a training data set by selecting, for each multidimensional point of the low-discrepancy sequence having one or more mapped samples, a mapped sample for inclusion in the training data set; and training one or more models used to generate autonomous driving decisions of an autonomous vehicle based on the selected training data set.

Abstract: Context-based data valuation and transmission, including: acquiring sensor data from a plurality of sensors of the autonomous vehicle, wherein the sensor data comprises a plurality of portions and the plurality of sensors comprises at least two cameras; determining, for each portion of the sensor data, a value based on one or more conditions under which the sensor data was captured; determining, based on the values for the sensor data, an upload policy; and transmitting, based on the upload policy, one or more portions of the sensor data to a server.

Abstract: Autonomous vehicle model training and validation using low-discrepancy sequences may include: generating a low-discrepancy sequence in a multidimensional space comprising a plurality of multidimensional points; mapping each sample of a plurality of samples of a data corpus to a corresponding entry in the low-discrepancy sequence, wherein each sample of the plurality of samples comprises one or more environmental descriptors for an environment relative to a vehicle and one or more state descriptors describing a state of the vehicle; selecting, from the data corpus, a training data set by selecting, for each multidimensional point of the low-discrepancy sequence having one or more mapped samples, a mapped sample for inclusion in the training data set; and training one or more models used to generate autonomous driving decisions of an autonomous vehicle based on the selected training data set.

Abstract: Indexing a data corpus to a set of multidimensional points, including: generating a set of points in a multidimensional space; identifying, for each sample in a plurality of samples in a data corpus, a nearest point in the set of points; and generating an index mapping each sample with the nearest point in the set of points.

Abstract: Modifying vehicle display parameters using operator gaze, including: detecting a gaze of an operator of a vehicle; and modifying one or more parameters of a display based on the gaze of the operator.

Abstract: Blended operator and autonomous control in an autonomous vehicle, including: receiving sensor data from a plurality of sensors of an autonomous vehicle; determining, based on the sensor data, a degree of autonomous control for each control input of a plurality of control inputs; and applying the degree of autonomous control for each control input of the plurality of control inputs.

Abstract: Shared memory access in a distributed system, including: receiving a memory access request associated with a time value; determining, based on the time value, an entry in a translation lookaside buffer (TLB); and determining, based on the entry, whether to allow the memory access request.

Abstract: Dynamic calibration of cameras in a stereoscopic configuration may include: determining a disparity between first image data from a first camera and second image data from a second camera, wherein the first camera and the second camera are in a stereoscopic configuration, and wherein the disparity comprises a difference in placement of one or more objects in the first image data relative to the second image data; and adjusting one or more of the first camera or the second camera, based on the disparity and sensor data from a sensor other than the first camera and the second camera, to calibrate the stereoscopic configuration of the first camera and the second camera to achieve stereoscopic camera distance functionality.

Abstract: Modifying settings of autonomous vehicle sensors based on predicted environmental states, including: determining, based on sensor data, that a predicted environmental state of the autonomous vehicle is associated with a sensory input outside an operating range of a sensor; determining a modified operating range of the sensor; and modifying one or more sensors to operate according to the modified operating range.

Abstract: Distributed data sampling, including: receiving a sampling target; generating, based on one or more sensors, sampled data; determining, based on the sampling target, a value for the sampled data; and determining, based on the value for the sampled data, whether to provide the sampled data to a remotely disposed computing device.