Tree based behavior predictor
Various embodiments include methods and devices for training and implementing a tree-based behavior prediction model for use in autonomous vehicle control systems. Some embodiments may include labeling real-world autonomous vehicle run data to indicate an insight of the data, selecting an insight decision tree of the tree-based behavior prediction model for training using the labeled data, training the insight decision tree using the labeled data to classify a probability of an insight associated with the insight decision tree, and updating the tree-based behavior prediction model based on training the insight decision tree. Some embodiments may include selecting an insight decision tree of a tree-based behavior prediction model configured for classifying a probability of an insight associated with the insight decision tree, executing the insight decision tree, and outputting a probability of an insight determined from executing the insight decision tree using the data.
1 . A method for training a tree-based behavior prediction model, comprising:
receiving real-world autonomous vehicle run data;
labeling the real-world autonomous vehicle run data at a plurality of time slices to generate labeled data indicating to a processing device an insight of the real-world autonomous vehicle run data;
processing the real-world autonomous vehicle run data to determine features of the real-world autonomous vehicle run data at the plurality of time slices;
associating the features and the real-world autonomous vehicle run data at one or more time slices of the plurality of time slices;
selecting an insight decision tree of the tree-based behavior prediction model for training using the labeled data and the determined associations, wherein the selection is based on a correspondence between features of the insight decision tree and the determined features of the real-world autonomous vehicle run data and is associated with an insight such that each node of the insight decision tree represents a probability of the insight associated with the insight decision tree based on the real-world autonomous vehicle run data;
determining, based on a comparison of a number of nodes of the insight decision tree to a node amount threshold, the number of nodes of the insight decision tree exceeds the node amount threshold;
removing one or more nodes of the insight decision tree to reduce the number of nodes below the node amount threshold to generate a pruned insight decision tree;
training the pruned insight decision tree using the labeled real-world autonomous vehicle run data to classify a probability of the insight associated with the pruned insight decision tree;
assigning a confidence score to the trained insight decision tree based on a number of confirmed predictions of the trained insight decision tree; and
updating the tree-based behavior prediction model based on training the pruned insight decision tree using the labeled real-world autonomous vehicle run data and the assignment of the confidence score.
2 . The method of claim 1 , wherein labeling the real-world autonomous vehicle run data at the plurality of time slices comprises:
labeling a first time slice of the plurality of time slices with a ground truth insight; and
labeling other time slices of the plurality of time slices preceding the first time slice.
3 . The method of claim 1 , further comprising:
determining a feature of the real-world autonomous vehicle run data; and
associating the real-world autonomous vehicle run data, the feature of the real-world autonomous vehicle run data, and a label of the real-world autonomous vehicle run data,
wherein selecting the insight decision tree of the tree-based behavior prediction model for training comprises selecting the insight decision tree configured for classifying the probability of the insight of the label based on the feature.
4 . A computing device, comprising a processing device configured with executable instructions to perform operations comprising:
receiving real-world autonomous vehicle run data;
labeling the real-world autonomous vehicle run data at a plurality of time slices to generate labeled data indicating to a processing device an insight of the real-world autonomous vehicle run data;
processing the real-world autonomous vehicle run data to determine features of the real-world autonomous vehicle run data at the plurality of time slices;
associating the features and the real-world autonomous vehicle run data at one or more time slices of the plurality of time slices;
selecting an insight decision tree of a tree-based behavior prediction model for training using the labeled data and the determined associations, wherein selection is based on a correspondence between features of the insight decision tree and the determined features of the real-world autonomous vehicle run data and is associated with an insight such that each node of the insight decision tree represents a probability of the insight associated with the insight decision tree based on the real-world autonomous vehicle run data;
determining, based on a comparison of a number of nodes of the insight decision tree to a node amount threshold, the number of nodes of the insight decision tree exceeds the node amount threshold;
removing one or more nodes of the insight decision tree to reduce the number of nodes below the node amount threshold to generate a pruned insight decision tree;
training the pruned insight decision tree using the labeled real-world autonomous vehicle run data to classify a probability of the insight associated with the pruned insight decision tree;
assigning a confidence score to the trained insight decision tree based on a number of confirmed predictions of the trained insight decision tree; and
updating the tree-based behavior prediction model based on training the pruned insight decision tree using the labeled real-world autonomous vehicle run data and the assignment of the confidence score.
5 . The computing device of claim 4 , wherein the processing device is further configured with executable instructions to perform operations such that labeling the real-world autonomous vehicle run data at the plurality of time slices comprises:
labeling a first time slice of the plurality of time slices with a ground truth insight; and
labeling other time slices of the plurality of time slices preceding the first time slice.
6 . The computing device of claim 4 , wherein the processing device is further configured with executable instructions to perform operations further comprising:
determining a feature of the real-world autonomous vehicle run data; and
associating the real-world autonomous vehicle run data, the feature of the real-world autonomous vehicle run data, and a label of the real-world autonomous vehicle run data,
wherein the processing device is configured with executable instructions to perform operations such that selecting the insight decision tree of the tree-based behavior prediction model for training comprises selecting the insight decision tree configured for classifying the probability of the insight of the label based on the feature.