Trained human-intention classifier for safe and efficient robot navigation

Patent metadata

Field	Value
Publication number	US-9776323-B2
Application number	US-201614989345-A
Country	US
Kind code	B2
Filing date	Jan 6, 2016
Priority date	Jan 6, 2016
Publication date	Oct 3, 2017
Grant date	Oct 3, 2017

Abstract

A trained classifier to be used with a navigation algorithm for use with mobile robots to compute safe and efficient trajectories. An offline learning process is used to train a classifier for the navigation algorithm (or motion planner), and the classifier functions, after training is complete, to accurately detect intentions of humans within a space shared with the robot to block the robot from traveling along its current trajectory. At runtime, the trained classifier can be used with regression based on past trajectories of humans (or other tracked, mobile entities) to predict where the humans will move in the future and whether the humans are likely to be blockers. The planning algorithm or motion planner generates trajectories based on predictions of human behavior that allow the robot to navigate amongst crowds of people more safely and efficiently.

First claim

We claim:

1. A mobile robot, comprising:
a drive system operating to move the mobile robot in a workspace;
a robot controller transmitting control signals to the drive system to follow a trajectory through the workspace;
a motion planner periodically updating the trajectory; and
a human-intention classifier generating a prediction of a future behavior for at least one mobile entity in the workspace,
wherein the motion planner performs the updating of the trajectory using the behavior predictions for the at least one mobile entity, and
wherein the behavior predictions comprise a blocking score for the at least one mobile entity defining a probability that the mobile entity will block travel of the mobile robot by intentionally interacting with the mobile robot.

2. The mobile robot of claim 1 , wherein the behavior involves a behavior that blocks future movements of the mobile robot along the trajectory and the behavior that blocks future movement includes intentional interaction with the mobile robot.

3. The mobile robot of claim 1 , wherein the updating of the trajectory involves moving toward one or more of the at least one mobile entity with a blocking score corresponding with non-blocking entities.

4. The mobile robot of claim 1 , wherein the human-intention classifier is trained to perform the prediction generation in at least one of an offline process completed prior to runtime of the mobile robot, or wherein the prediction generating and the updating of the trajectory are performed during runtime of the mobile robot.

5. The mobile robot of claim 4 , wherein the offline process comprises:
gathering, while a robot moves along a goal trajectory through a space shared with a plurality of humans, a set of trajectories for the humans;
identifying a subset of the gathered trajectories for the humans as interfering with movement of the robot along the goal trajectory; and
assigning a blocking score indicating each of the subset of the gathered trajectories is associated with one of the humans that blocks the mobile robot.

6. The mobile robot of claim 5 , wherein the prediction generation includes receiving a past trajectory for one of the at least one mobile entity in the workspace and comparing the past trajectory with the subset of the gathered trajectories from the offline process to generate a blocking score for the one of the at least one mobile entity associated with the past trajectory, whereby the past trajectory is used to provide the behavior prediction.

7. The mobile robot of claim 1 , wherein the motion planner computes a repulsive potential for each of the at least one mobile entity based on the behavior prediction and generates a predicted trajectory in the workspace for each of the at least one mobile entity using trajectory regression.

8. The mobile robot of claim 7 , wherein the motion planner performs the updating of the trajectory based on the repulsive potentials and the predicted trajectories.

9. A method of creating a human-intention classifier for a mobile robot, comprising:
for a predefined time period, with a robot controller, controlling a robot to move along a goal trajectory in a space shared with a plurality of humans;
during the predefined time period, with the robot controller, recording trajectories of the robot and the humans; and
with the robot controller, classifying each of the recorded trajectories for the humans as being associated either with blocking of the robot on the goal trajectory or non-blocking of the travel of the robot on the goal trajectory, wherein the blocking of the robot includes intentional interaction with the robot and wherein the non-blocking of the travel of the robot includes cooperative behavior to avoid collision.

10. The method of claim 9 , further including assigning a first blocking score to each of the recorded trajectories associated with the blocking and a second blocking score, differing from the first blocking score, to each of the recorded trajectories associated with the non-blocking of the travel of the robot.

11. The method of claim 10 , wherein the recorded trajectories for the humans are associated with the blocking of the robot when the recorded trajectories correspond with a modification of a trajectory of the robot from the goal trajectory.

12. The method of claim 9 , further comprising extracting from the recorded trajectories at least one variable from the group consisting of absolute position in the space, velocity, and acceleration, whereby the extracted variable is associated with the recorded trajectory for use in comparing tracked trajectories of a human to provide predictions of future behavior by the human.

13. The method of claim 9 , further comprising extracting, relative to the recorded trajectories for the robot, from the recorded trajectories for the humans at least one variable from the group consisting of relative position in the space, relative velocity, and relative acceleration, whereby the extracted variable is associated with the recorded trajectory for use in comparing tracked trajectories of a human to provide predictions of future behavior by the human.

14. A method of navigating a robot through a space shared with one or more mobile humans, comprising:
positioning a mobile robot in a space;
first controlling, with a robot controller, the mobile robot to move along a first trajectory toward a goal location in the space;
determining a past trajectory of a human in the space;
assigning a blocking score to the past trajectory for the human by comparing the past trajectory to a set of prerecorded trajectories each being pre-classified with a probability of blocking behavior, wherein the blocking behavior includes intentional interaction with the mobile robot by the human; and
second controlling, with the robot controller, the mobile robot to move along either the first trajectory or along a second trajectory toward the goal location in the space, wherein the second trajectory is computed based on the blocking score and wherein the second trajectory is computed to avoid the human in the space when the blocking score is greater than a first predefined value and to move toward the human in the space when the blocking score is less than a second predefined value.

15. The method of claim 14 , wherein the set of prerecorded trajectories are associated with a first set of trajectories for humans moving within a test space identified as blocking travel of a test robot concurrently moving in the test space and with a second set of trajectories for humans in the test space identified as not blocking travel of the test robot.

16. The method of claim 15 , wherein the past trajectory includes at least one of variable in the group consisting of absolute position, velocity, acceleration, position relative to the mobile robot, velocity relative to the mobile robot, and acceleration relative to the mobile robot.

17. The method of claim 16 , wherein a value of the at least one variable is compared with a value of a matching variable extracted from the trajectories in the first and second sets.

18. The method of claim 14 , wherein the second trajectory is computed based on a repulsive potential value assigned to the human and wherein the repulsive potential value being computed based on the blocking score.

Assignees

• Disney Entpr Inc

Inventors

• O'Sullivan Carol Ann
• Park Chonhyon
• Gilbert Max L
• Ondrej Jan
• Freeman Kyle G

Classifications

• G06N20/00

  Machine learning · CPC title

• Y10S901/46

  Sensing device · CPC title

• B25J9/1666 Primary

  Avoiding collision or forbidden zones · CPC title

• Y10S901/01

  Mobile robot · CPC title

• B25J9/163

  learning, adaptive, model based, rule based expert control · CPC title