Real time decision making for autonomous driving vehicles
In one embodiment, a method, apparatus, and system may predict behavior of environmental objects using machine learning at an autonomous driving vehicle (ADV). One or more yield/overtake decisions are made with respect to one or more objects in the ADV's surrounding environment using a data processing architecture comprising at least a first, a second, and a third neural networks, the first, the second, and the third neural networks having been trained with a training data set. Driving signals are generated based at least in part on the yield/overtake decisions to control operations of the ADV.
1. A computer-implemented method for making a decision in operating an autonomous driving vehicle (ADV) using machine learning, comprising:
making one or more yield/overtake decisions with respect to one or more objects in the ADV's surrounding environment using a data processing architecture comprising at least a first, a second, and a third neural networks, the first, the second, and the third neural networks having been trained with a training data set, wherein the second neural network generates data encoding both extracted historical features of the one or more objects and map information as outputs; and
generating driving signals based at least in part on the one or more yield/overtake decisions to control operations of the ADV.
2. The method of claim 1 , wherein the first neural network is a multilayer perceptron (MLP), wherein the second neural network is a convolutional neural network (CNN), and wherein the third neural network is a fully-connected network.
3. The method of claim 2 , wherein the first neural network receives historical features of the one or more objects from one or more previous planning cycles as inputs, and generates the extracted historical features of the one or more objects as outputs.
4. The method of claim 3 , wherein the second neural network receives the extracted historical features of the one or more objects and the map information as inputs.
5. The method of claim 4 , wherein the third neural network receives the encoded data and historical features of the ADV as inputs, and generates the one or more yield/overtake decisions comprising decisions with respect to each of the one or more objects as outputs.
6. The method of claim 5 , wherein the encoded data and the historical features of the ADV are concatenated before being fed into the third neural network.
7. The method of claim 4 , wherein the historical features of the one or more objects comprise one or more of: a position, a speed, or an acceleration, and wherein the map information is derived from a high-definition map and comprises one or more of: a lane feature component, a traffic signal component, a static object component, or a general map information component.
8. The method of claim 4 , wherein the extracted historical features of the one or more objects and the map information are labeled with associated block information based on a grid subdivision of a rectangular perception area of the ADV, the grid subdivision comprising subdividing the rectangular perception area of the ADV into a plurality of uniformly sized rectangular blocks based on a grid.
9. The method of claim 1 , wherein the training data set comprises previously recorded driving and perception data automatically labeled with yield/overtake decisions.
10. A non-transitory machine-readable medium having instructions stored therein, which when executed by a processor, cause the processor to perform operations for making a decision in operating an autonomous driving vehicle (ADV) using machine learning, the operations comprising:
making one or more yield/overtake decisions with respect to one or more objects in the ADV's surrounding environment using a data processing architecture comprising at least a first, a second, and a third neural networks, the first, the second, and the third neural networks having been trained with a training data set, wherein the second neural network generates data encoding both extracted historical features of the one or more objects and map information as outputs; and
generating driving signals based at least in part on the yield/overtake decisions to control operations of the ADV.
11. The non-transitory machine-readable medium of claim 10 , wherein the first neural network is a multilayer perceptron, wherein the second neural network is a convolutional neural network (CNN), and wherein the third neural network is a fully-connected network.
12. The non-transitory machine-readable medium of claim 11 , wherein the first neural network receives historical features of the one or more objects from one or more previous planning cycles as inputs, and generates the extracted historical features of the one or more objects as outputs.
13. The non-transitory machine-readable medium of claim 12 , wherein the second neural network receives the extracted historical features of the one or more objects and the map information as inputs.
14. The non-transitory machine-readable medium of claim 13 , wherein the third neural network receives the encoded data and historical features of the ADV as inputs, and generates the one or more yield/overtake decisions comprising decisions with respect to each of the one or more objects as outputs.
15. The non-transitory machine-readable medium of claim 14 , wherein the encoded data and the historical features of the ADV are concatenated before being fed into the third neural network.
16. The non-transitory machine-readable medium of claim 13 , wherein the historical features of the one or more objects comprise one or more of: a position, a speed, or an acceleration, and wherein the map information is derived from a high-definition map and comprises one or more of: a lane feature component, a traffic signal component, a static object component, or a general map information component.
17. The non-transitory machine-readable medium of claim 13 , wherein the extracted historical features of the one or more objects and the map information are labeled with associated block information based on a grid subdivision of a rectangular perception area of the ADV, the grid subdivision comprising subdividing the rectangular perception area of the ADV into a plurality of uniformly sized rectangular blocks based on a grid.
18. The non-transitory machine-readable medium of claim 10 , wherein the training data set comprises previously recorded driving and perception data automatically labeled with yield/overtake decisions.
19. A data processing system, comprising:
a processor; and
a memory coupled to the processor to store instructions, which when executed by the processor, cause the processor to perform operations for making a decision in operating an autonomous driving vehicle (ADV) using machine learning, the operations including
making one or more yield/overtake decisions with respect to one or more objects in the ADV's surrounding environment using a data processing architecture comprising at least a first, a second, and a third neural networks, the first, the second, and the third neural networks having been trained with a training data set, wherein the second neural network generates data encoding both extracted historical features of the one or more objects and map information as outputs, and
generating driving signals based at least in part on the yield/overtake decisions to control operations of the ADV.
20. The data processing system of claim 19 , wherein the first neural network is a multilayer perceptron, wherein the second neural network is a convolutional neural network (CNN), and wherein the third neural network is a fully-connected network.
21. The data processing system of claim 20 , wherein the first neural network receives historical features of the one or more objects from one or more previous planning cycles as inputs, and generates the extracted historical features of the one or more objects as outputs.
22. The data processing system of claim 21 , wherein the second neural network receives the extracted historical features of the one or more objects and the map information as inputs.
23. The data processing system of claim 22 , wherein the third neural network receives the encoded data and historical features of the ADV as inputs, and generates the one or more yield/overtake decisions comprising decisions with respect to each of the one or more objects as outputs.