Systems and methods for predicting entity behavior
Systems and method are provided for controlling a vehicle. In one embodiment, a method includes: receiving sensor data sensed from an environment associated with the vehicle; processing, by a processor, the sensor data to determine observation data, the observation data including differential features associated with an agent in the environment; determining, by the processor, a context associated with the agent based on the observation; selecting, by the processor, a first probability model associated with the context; processing, by the processor, the observation data with the selected first probability model to determine a set of predictions; processing, by the processor, the set of predictions with a second probability model to determine a final prediction of interaction behavior associated with the agent; and selectively controlling, by the processor, the vehicle based on the final prediction of interaction behavior associated with the agent.
1. A method of controlling a vehicle, comprising:
receiving sensor data sensed from an environment associated with the vehicle;
processing, by a processor, the sensor data to determine observation data, the observation data including differential features associated with an agent in the environment;
determining, by the processor, a context associated with the agent based on the observation data;
selecting, by the processor, a first probability model associated with the context from a plurality of probability models, wherein each of the plurality of probability models are trained Gaussian Mixture Model-Hidden Markov Models for an associated context;
processing, by the processor, the observation data with the selected first probability model to determine a set of predictions;
processing, by the processor, the set of predictions with a second probability model to determine a final prediction of interaction behavior associated with the agent, wherein the second probability model is a trained sparsely correlated Hidden Markov Model; and
selectively controlling, by the processor, the vehicle based on the final prediction of interaction behavior associated with the agent.
2. The method of claim 1 , wherein the differential features are associated with an acceleration and a heading.
3. The method of claim 1 , wherein the differential features include an x and y location.
4. The method of claim 3 , wherein the context is determined from the x and y location.
5. The method of claim 1 , wherein the context is associated with a lane in which the agent is travelling.
6. The method of claim 5 , wherein the context is determined by using a sorted pairwise distance between an x and y location of the agent to a centerline of the lane, and mapping the lane to the context.
7. The method of claim 1 , further comprising performing unsupervised training of the Gaussian Mixture Model-Hidden Markov Models based on the context.
8. The method of claim 1 , further comprising performing unsupervised training of the sparsely correlated Hidden Markov Model based on the context.
9. A system for controlling a vehicle, comprising:
a sensor system configured to observe an agent in an environment associated with the vehicle and to produce sensor data based thereon; and
a prediction module configured to, by a processor, receive the sensor data, process the sensor data to determine observation data including differential features associated with an agent, determine a context associated with the agent based on the observation data, select a first probability model associated with the context from a plurality of probability models, wherein each of the plurality of probability models is a trained Gaussian Mixture Model-Hidden Markov Model for an associated context, process the observation data with the selected first probability model to determine a set of predictions, process the set of predictions with a second probability model to determine a final prediction of interaction behavior associated with the agent, wherein the second probability model is a trained sparsely correlated Hidden Markov Model, and selectively control the vehicle based on the final prediction of interaction behavior associated with the agent.
10. The system of claim 9 , wherein the differential features are associated with an acceleration and a heading.
11. The system of claim 9 , wherein the differential features include an x and y location.
12. The system of claim 11 , wherein the context is determined from the x and y location.
13. The system of claim 9 , wherein the context is associated with a lane in which the agent is travelling.
14. The system of claim 13 , wherein the context is determined by using a sorted pairwise distance between an x and y location of the agent to a centerline of the lane, and mapping the lane to the context.
15. The system of claim 9 , wherein the prediction module performs unsupervised training of the Gaussian Mixture Model-Hidden Markov Model based on the context.
16. The system of claim 9 , wherein the prediction module performs unsupervised training of the sparsely correlated Hidden Markov Model based on the context.