System and method for early event detection using generative and discriminative machine learning models

1. A method for human-robot collaboration, the method comprising:

acquiring visual temporal data of a human partner to a robot;

determining, using a generative module, predicted future visual temporal data in response to the visual temporal data, the visual temporal data including current visual temporal data and previous visual temporal data; and

determining, using a discriminative module, a vector of probabilities indicating the likelihood that a future action of the human partner belongs to each class among a set of classes being considered in response to at least the future visual temporal data and the visual temporal data;

modifying, using a predictive module, the visual temporal data by concatenating the predicted future visual temporal data to the visual temporal data, and inputting the visual temporal data that has been modified by the predictive module to the discriminative module;

determining, using a transition enforcement module, a predicted class of the set of classes by combining the vector of probabilities and a transition matrix, the transition matrix containing conditional probabilities of the future action of the human partner taking place based upon the visual temporal data.

2. The method of claim 1 , further comprising:

determining, using a robot planning module, a robot action that best suits the visual temporal data and the predicted class.

3. The method of claim 2 , further comprising:

actuating the robot in accordance with the robot action.

4. The method of claim 1 , further comprising:

capturing, using a data acquisition module, visual temporal data.

5. The method of claim 4 , wherein the data acquisition module comprises at least one of an external depth sensor and a Red-Green-Blue sensor.

6. The method of claim 1 , wherein the visual temporal data comprises at least one of a Red-Green-Blue video, an infrared video, a near-infrared video, and depth map sequences.

7. An apparatus for human-robot collaboration, the apparatus comprising:

a controller comprising:

a processor; and

a memory comprising computer-executable instructions that, when executed by the processor, cause the processor to perform operations, the operations comprising:

acquiring visual temporal data of a human partner to a robot;

determining, using a generative module, predicted future visual temporal data in response to the visual temporal data, the visual temporal data including current visual temporal data and previous visual temporal data; and

determining, using a discriminative module, a vector of probabilities indicating the likelihood that a future action of the human partner belongs to each class among a set of classes being considered in response to at least the future visual temporal data and the visual temporal data;

modifying, using a predictive module, the visual temporal data by concatenating the predicted future visual temporal data to the visual temporal data, and inputting the visual temporal data that has been modified by the predictive module to the discriminative module;

determining, using a transition enforcement module, a predicted class of the set of classes by combining the vector of probabilities and a transition matrix, the transition matrix containing conditional probabilities of the future action of the human partner taking place based upon the visual temporal data.

8. The apparatus of claim 7 , wherein the operations further comprise:

determining, using a robot planning module, a robot action that best suits the visual temporal data and the predicted class.

9. The apparatus of claim 8 , wherein the operations further comprise:

actuating the robot in accordance with the robot action.

10. The apparatus of claim 7 , further comprising:

a data acquisition module configured to capture the visual temporal data.

11. The apparatus of claim 10 , wherein the data acquisition module comprises at least one of an external depth sensor and a Red-Green-Blue sensor.

12. The apparatus of claim 7 , wherein the visual temporal data comprises at least one of a Red-Green-Blue video, an infrared video, a near-infrared video, and depth map sequences.

13. A computer program product embodied on a non-transitory computer readable medium, the computer program product including instructions that, when executed by a processor, cause the processor to perform operations comprising:

acquiring visual temporal data of a human partner to a robot;

determining, using a generative module, predicted future visual temporal data in response to the visual temporal data, the visual temporal data including current visual temporal data and previous visual temporal data; and

determining, using a discriminative module, a vector of probabilities indicating the likelihood that a future action of the human partner belongs to each class among a set of classes being considered in response to at least the future visual temporal data and the visual temporal data;

modifying, using a predictive module, the visual temporal data by concatenating the predicted future visual temporal data to the visual temporal data, and inputting the visual temporal data that has been modified by the predictive module to the discriminative module;

determining, using a transition enforcement module, a predicted class of the set of classes by combining the vector of probabilities and a transition matrix, the transition matrix containing conditional probabilities of the future action of the human partner taking place based upon the visual temporal data.

14. The computer program product of claim 13 , wherein the operations further comprise:

determining, using a robot planning module, a robot action that best suits the visual temporal data and the predicted class.

15. The computer program product of claim 14 , wherein the operations further comprise:

actuating the robot in accordance with the robot action.