SYSTEMS AND METHODS FOR MULTI-OBJECT NAVIGATION IN REAL ENVIRONMENTS
A navigating robot includes: a feature module configured to detect objects in images captured by a camera of the navigating robot while the navigating robot is in a real world space; a mapping module configured to generate a map including locations of objects captured in the images and at least one attribute of the objects; and a navigation module trained to find and navigate to N different objects in the real world space in a predetermined order by: when a location of a next one of the N different objects in the predetermined order is stored in the map, navigate toward the next one of the N different objects in the real world space; and when the location of the next one of the N different objects in the predetermined order is not stored in the map, navigate to a portion of the map not yet captured in any images.
1 . A navigating robot, comprising:
a feature module configured to detect objects in images captured by a camera of the navigating robot while the navigating robot is in a real world space;
a mapping module configured to generate a map including locations of objects captured in the images and at least one attribute of the objects; and
a navigation module trained to find and navigate to N different objects in the real world space in a predetermined order by:
when a location of a next one of the N different objects in the predetermined order is stored in the map, actuate one or more propulsion devices of the navigating robot and navigate toward the next one of the N different objects in the real world space; and
when the location of the next one of the N different objects in the predetermined order is not stored in the map, actuate one or more propulsion devices of the navigating robot and navigate to a portion of the real world space in the map not yet captured in any images from the camera.
2 . The navigating robot of claim 1 wherein the navigation module is trained based on finding and navigating to N simulated objects in a predetermined order in a simulated space.
3 . The navigating robot of claim 1 wherein the mapping module is configured to generate the map further based on input from a light detection and ranging (LIDAR) sensor of the navigating robot.
4 . The navigating robot of claim 1 wherein the navigation module is configured to, when the location of the next one of the N different objects in the predetermined order is stored in the map:
determine a path to the location of the next one of the N different objects; and
actuate the one or more propulsion devices of the navigating robot and navigate along the path toward the next one of the N different objects in the real world space.
5 . The navigating robot of claim 4 wherein the path is a shortest path.
6 . The navigating robot of claim 1 wherein at least a portion of the navigation module is trained using reinforcement learning.
7 . The navigating robot of claim 1 wherein the navigation module is configured to, when the next one of the N different objects in the predetermined order is captured in an image in route to the portion of the real world space, actuate the one or more propulsion devices of the navigating robot and navigate toward the next one of the N different objects in the real world space.
8 . The navigating robot of claim 1 wherein the N different objects are all different types of object.
9 . The navigating robot of claim 1 wherein the N different objects are different colors.
10 . The navigating robot of claim 1 wherein the N different objects are visually different than each other one of the N different objects.
11 . The navigating robot of claim 1 wherein the navigation module includes:
an encoder module configured to receive the map and generate an encoding of the map;
a long short term memory (LSTM) module configured to generate an output based on the encoding;
a decoder module configured to decode the output of the LSTM module into a second map;
a masking module configured to apply a mask to the second map; and
a sampling module configured to sample the portion of the real world space in the map not yet captured in any images from the camera from the second map.
12 . The navigating robot of claim 11 wherein the mask is a fog-of-war mask.
13 . The navigating robot of claim 11 wherein the sampling module is configured to sample the portion from the second map based on maximizing a reward.
14 . A navigating robot, comprising:
a means for detecting objects in images captured by a camera of the navigating robot while the navigating robot is in a real world space;
a means for generating a map including locations of objects captured in the images and at least one attribute of the objects; and
a means trained to find and navigate to N different objects in the real world space in a predetermined order by:
when a location of a next one of the N different objects in the predetermined order is stored in the map, actuate one or more propulsion devices of the navigating robot and navigate toward the next one of the N different objects in the real world space; and
when the location of the next one of the N different objects in the predetermined order is not stored in the map, actuate one or more propulsion devices of the navigating robot and navigate to a portion of the real world space in the map not yet captured in any images from the camera.
15 . A navigation method, comprising:
detecting objects in images captured by a camera of a navigating robot while the navigating robot is in a real world space;
generating a map including locations of objects captured in the images and at least one attribute of the objects; and
based on training, navigating to N different objects in the real world space in a predetermined order by:
when a location of a next one of the N different objects in the predetermined order is stored in the map, actuating one or more propulsion devices of the navigating robot and navigating toward the next one of the N different objects in the real world space; and
when the location of the next one of the N different objects in the predetermined order is not stored in the map, actuating one or more propulsion devices of the navigating robot and navigating to a portion of the real world space in the map not yet captured in any images from the camera.
16 . The navigation method of claim 15 wherein the training includes training based on finding and navigating to N simulated objects in a predetermined order in a simulated space.
17 . The navigation method of claim 15 wherein generating the map includes generating the map further based on input from a light detection and ranging (LIDAR) sensor of the navigating robot.
18 . The navigation method of claim 15 wherein, when the location of the next one of the N different objects in the predetermined order is stored in the map, the navigation method includes:
determining a path to the location of the next one of the N different objects; and
actuating the one or more propulsion devices of the navigating robot and navigating along the path toward the next one of the N different objects in the real world space.
19 . The navigation method of claim 15 wherein at least a portion of the training includes training using reinforcement learning.
20 . The navigation method of claim 15 wherein, when the next one of the N different objects in the predetermined order is captured in an image in route to the portion of the real world space, the navigation method includes actuating the one or more propulsion devices of the navigating robot and navigating toward the next one of the N different objects in the real world space.
21 . The navigation method of claim 15 wherein at least one of:
the N different objects are all different types of object;
the N different objects are different colors; and
the N different objects are visually different than each other one of the N different objects.
22 . The navigation method of claim 15 further comprising:
receiving the map and generating an encoding of the map;
by a long short term memory (LSTM) module, generating an output based on the encoding;
decoding the output into a second map;
applying a mask to the second map; and
sampling the portion of the real world space in the map not yet captured in any images from the camera from the second map.
23 . The navigation method of claim 22 wherein the sampling includes sampling the portion from the second map based on maximizing a reward.