SYSTEM AND METHOD FOR FLEET SCENE INQUIRIES
Systems and methods for fleet scene inquiries are disclosed herein. In one aspect, an autonomous vehicle includes at least one sensor configured to output sensor data, a memory, and a network communications subsystem configured to receive a scene capture inquiry from an oversight system. The autonomous vehicle further includes a processor configured to: compare the sensor data to the scene capture inquiry based at least in part on a deep learning model, and in response to the sensor data at least generally matching the scene capture inquiry, store the sensor data in the memory as scene data. The deep learning model is trained based on the scene data.
1 . An autonomous vehicle, comprising:
at least one sensor configured to output sensor data;
a memory;
a network communications subsystem configured to receive a scene capture inquiry from an oversight system; and
a processor configured to:
compare the sensor data to the scene capture inquiry based at least in part on a deep learning model, and
in response to the sensor data at least generally matching the scene capture inquiry, store the sensor data in the memory as scene data,
wherein the deep learning model is trained based on the scene data.
2 . The autonomous vehicle of claim 1 , wherein the scene capture inquiry comprises one or more parametric values that define one or more types of the sensor data to be stored as scene data.
3 . The autonomous vehicle of claim 2 , wherein the one or more parametric values comprises one or more of the following: a time of day parameter, a geofencing parameter, a road condition parameter, and a weather parameter.
4 . The autonomous vehicle of claim 2 , wherein the processor is further configured to compare the sensor data to the one or more parametric values on a frame-by-frame basis.
5 . The autonomous vehicle of claim 1 , wherein the scene capture inquiry is generated by the oversight system based on an input that indicates that the deep learning model has low confidence for a scene defined by the scene capture inquiry.
6 . The autonomous vehicle of claim 1 , wherein the deep learning model is used as part of a perception algorithm running on the autonomous vehicle, the perception algorithm configured to detect objects within the sensor data.
7 . The autonomous vehicle of claim 1 , further comprising:
a vehicle drive subsystem configured to drive the autonomous vehicle based at least in part on the sensor data.
8 . The autonomous vehicle of claim 1 , wherein:
the processor is further configured to in response to the scene data meeting an offloading condition, transmit the scene data to an offboard data storage, and
the offloading condition comprises a threshold amount of scene data stored on the memory.
9 . The autonomous vehicle of claim 1 , wherein:
the processor is further configured to in response to the scene data meeting an offloading condition, transmit the scene data to an offboard data storage, and
the offloading condition comprises a threshold amount of time elapsed since the network communications subsystem received the scene capture inquiry.
10 . The autonomous vehicle of claim 1 , wherein the offboard data storage is a cloud server.
11 . A method performed by a processor of an autonomous vehicle, comprising:
receiving sensor data output from a plurality of sensors of the autonomous vehicle;
receiving a scene capture inquiry from an oversight system via a network communications subsystem of the autonomous vehicle;
comparing the sensor data to the scene capture inquiry based at least in part on a deep learning model; and
in response to the sensor data at least generally matching the scene capture inquiry, storing the sensor data in a memory of the autonomous vehicle as scene data,
wherein the deep learning model is trained based on the scene data.
12 . The method of claim 11 , wherein the scene data comprises a collection of data received from two or more of the sensors substantially simultaneously.
13 . The method of claim 11 , further comprising:
receiving a perception algorithm comprising the trained deep learning model; and
running the perception algorithm at the autonomous vehicle.
14 . The method of claim 13 , further comprising:
driving the autonomous vehicle based at least in part on the sensor data and the perception algorithm.
15 . The method of claim 13 , wherein the perception algorithm is configured to detect objects within the sensor data.
16 . A system for training a deep learning model, comprising:
an oversight system configured to generate a scene capture inquiry;
an offboard data storage; and
a plurality of autonomous vehicles, each of the autonomous vehicles comprising:
a vehicle sensor subsystem comprising a plurality of sensors configured to output sensor data,
a memory,
a network communications subsystem configured to receive the scene capture inquiry from the oversight system, and
a processor configured to:
compare the sensor data to the scene capture inquiry based at least in part on a deep learning model, and
in response to the sensor data at least generally matching the scene capture inquiry, store the sensor data in the memory as scene data,
wherein the deep learning model is trained based on the scene data.
17 . The system of claim 16 , wherein the oversight system is configured to receive an input from a user identifying a type of scene to be obtained by the autonomous vehicles, wherein the generation of the scene capture inquiry is based on the input received from the user.
18 . The system of claim 16 , wherein the processor of each of the autonomous vehicles is further configured to:
run a perception algorithm configured to identify objects in the sensor data for the corresponding autonomous vehicle; and
generate a confidence value indicative of a confidence that objects detected by the perception algorithm are accurate based on one or more parameters defining environmental conditions when the sensor data was obtained.
19 . The system of claim 18 , wherein the oversight system is configured to receive an input from one or more of the autonomous vehicles that indicates that the perception algorithm has a low confidence for the corresponding environmental conditions.
20 . The system of claim 16 , wherein the oversight system is further configured to receive input from a user indicative of whether the scene data stored in the offboard data storage correctly matches the scene capture inquiry prior to using the scene data for training the deep learning model.