Edge enhanced incremental learning for autonomous driving vehicles
A system comprises a computer including a processor and a memory. The memory includes instructions such that the processor is programmed to: determine, via a trained neural network model, a route for a vehicle to traverse based on vehicle sensor data, and update the trained neural network model based on data received from at least one of an edge computing device or an infrastructure (V2I) device.
1. A system for edge enhanced incremental learning for autonomous driving vehicles, the system comprising:
a computer including a processor and a memory, the memory including instructions such that the processor is programmed to:
determine, via a trained neural network model, a route for a vehicle to traverse based on vehicle sensor data; and
update the trained neural network model based on data received from at least one of an edge computing device or an infrastructure (V2I) device,
wherein the edge computing device comprises an edge server,
wherein the processor is further programmed to store a data structure representing a change in weights between the trained neural network model and the updated trained neural network model, wherein the processor is further programmed to upload the data structure to a specification platform, wherein the specification platform retrains the trained neural network model with stored data according to:
N=M′⊕Σ i n Δwi,
where N represents an updated neural network model, M′ represents a neural network model retained at the specification platform, ⊕ represents a weight update operator, Σ represents a summation of a weight operation applied to a difference in weights between neural network models ranging from i to n, wherein the difference in weights is maintained within records that include the differences in weights for each incrementally updated neural network model; and
selectively performing autonomous driving functions including: braking, acceleration, and steering of the vehicle via actuators implemented by control signals from the computer, unless a difference in neural network model accuracy is greater than a non-accurate parameter set by empirical analysis based on various environments encountered by autonomous vehicles, wherein when the difference in neural network model accuracy is greater than the non-accurate parameter, the system causes the vehicle to convert from an autonomous mode of operation to a manual mode of operation.
2. The system of claim 1 , wherein the processor is further programmed to update the trained neural network model based on data received from at least one other vehicle.
3. The system of claim 1 , wherein the processor is further programmed to receive the trained neural network model via an over-the-air (OTA) update.
4. The system of claim 1 , wherein the processor is further programmed to determine another route via the updated trained neural network model.
5. The system of claim 1 , wherein the processor is further programmed to cause the vehicle to traverse the route.
6. The system of claim 1 , wherein the infrastructure device comprises a roadside device.
7. The system of claim 1 , wherein the processor is further programmed to receive a second trained neural network model from the specification platform, wherein the second trained neural model includes updated weights based on data stored within the data structure.
8. A method for edge enhanced incremental learning for autonomous driving vehicles, the method comprising:
determining, via a trained neural network model, a route for a vehicle to traverse based on vehicle sensor data; and
updating the trained neural network model based on data received from at least one of an edge computing device or an infrastructure (V2I) device, wherein the edge computing device comprises an edge server;
storing, by a processor, a data structure representing a change in weights between the trained neural network model and the updated trained neural network model;
uploading, by the processor, the data structure to a specification platform;
retraining, by the specification platform, the trained neural network model with stored data according to:
N=M′⊕Σ i n Δwi,
where N represents an updated neural network model, M′ represents a neural network model retained at the specification platform, ⊕ represents a weight update operator, Σ represents a summation of a weight operation applied to a difference in weights between neural network models ranging from i to n, wherein the difference in weights is maintained within records that include the differences in weights for each incrementally updated neural network model; and
selectively performing autonomous driving functions including: braking, acceleration, and steering of the vehicle via actuators implemented by control signals from the computer, unless a difference in neural network model accuracy is greater than a non-accurate parameter set by empirical analysis based on various environments encountered by autonomous vehicles; and
when the difference in neural network model accuracy is greater than the non-accurate parameter, causing the vehicle to convert from an autonomous mode of operation to a manual mode of operation.
9. The method of claim 8 , the method further comprising updating the trained neural network model based on data received from at least one other vehicle.
10. The method of claim 8 , the method further comprising receiving the trained neural network model via an over-the-air (OTA) update.
11. The method of claim 8 , the method further comprising determining another route via the updated trained neural network model.
12. The method of claim 8 , the method further comprising causing the vehicle to traverse the route.
13. The method of claim 8 , wherein the infrastructure device comprises a roadside device.
14. The method of claim 8 , the method further comprising receiving a second trained neural network model from the specification platform, wherein the second trained neural model includes updated weights based on data stored within the data structure.