Apparatus and methods for training path navigation by robots
An apparatus and methods for training and/or operating a robotic device to follow a trajectory. A robotic vehicle may utilize a camera and stores the sequence of images of a visual scene seen when following a trajectory during training in an ordered buffer. Motor commands associated with a given image may be stored. During autonomous operation, an acquired image may be compared with one or more images from the training buffer in order to determine the most likely match. An evaluation may be performed in order to determine if the image may correspond to a shifted (e.g., left/right) version of a stored image as previously observed. If the new image is shifted left, right turn command may be issued. If the new image is shifted right then left turn command may be issued.
1. A method for operating a robot comprising:
operating in a first mode in which the robot executes actions in accordance with a training input and associates a first sensory context of a plurality of sensory contexts with a first behavior of a plurality of behaviors, the plurality of sensory contexts detected by at least one of a plurality of sensors of the robot;
configuring a predictor to receive the plurality of sensory contexts and to determine one or more of the plurality of behaviors of the robot based at least in part on detection of the plurality of sensory contexts, wherein the predictor outputs a control signal based at least in part on priorities assigned to each of the plurality of behaviors for each of the plurality of sensory contexts; and
operating in a second mode in which the robot executes actions based at least in part on an association made while the robot operated in the first mode, wherein:
the robot detects: (1) the first sensory context with a first sensor of the plurality of sensors, and (2) a second sensory context of the plurality of sensory contexts with a second sensor; and
the robot performs the first behavior based at least in part on the control signal outputted by the predictor upon receiving the first sensory context and the second sensory context.
2. The method of claim 1 , wherein the predictor determines the control signal using a winner-take-all computation.
3. The method of claim 1 , further comprising receiving a correction input from a user, the correction input altering a priority of a behavior assigned by the predictor in a sensory context.
4. The method of claim 3 , receiving the correction input by a combiner of the robot, and altering the priorities of the predictor using the combiner.
5. The method of claim 1 , wherein:
the first mode comprises a plurality of training iterations configured based on the training input; and
a given iteration is characterized by a control command and a performance measure associated with the actions execution based on the control command.
6. The method of claim 1 , adapting internal parameters based at least in part on a supervised learning using the predictor.
7. The method of claim 1 , providing a first sensor signal configured to convey an image frame comprising a plurality of pixels using the first sensor.
8. The method of claim 1 , providing a binary sensor signal characterized by one of two states using the second sensor.
9. The method of claim 1 , wherein the first sensory context comprises an obstacle detected by the first sensor.
10. The method of claim 1 , wherein the first behavior comprises following an obstacle avoidance trajectory.
11. A robot comprising:
a platform comprising a motor;
a plurality of sensors comprising a first sensor and a second sensor;
a controller configured to:
operate the robot in a first mode in which the robot executes actions in accordance with a training input and associates a first sensory context of a plurality of sensory contexts with a first behavior of a plurality of behaviors, the plurality of sensory contexts detected by at least one of a plurality of sensors of the robot;
receive the plurality of sensory contexts and determine one or more of the plurality of behaviors of the robot based at least in part on detection of the plurality of sensory contexts;
output a control signal based at least in part on priorities assigned to each of the plurality of behaviors for each of the plurality of sensory contexts; and
operate the robot in a second mode in which the robot executes actions based at least in part on an association made while the robot operated in the first mode, wherein:
the robot detects: (1) the first sensory context with a first sensor of the plurality of sensors, and (2) a second sensory context of the plurality of sensory contexts with a second sensor; and
the robot performs the first behavior based at least in part on the control signal outputted by the predictor upon receiving the first sensory context and the second sensory context.
12. The robot of claim 11 , wherein the first behavior comprises obstacle avoidance.
13. The robot of claim 11 , wherein the robot comprises a robotic vehicle.
14. The robot of claim 11 , wherein the first sensor comprises a camera.
15. The robot of claim 11 , wherein the second sensor comprises a LiDAR.
16. The robot of claim 11 , wherein the first sensory context comprises an obstacle detected by the first sensor.
17. A non-transitory computer-readable storage apparatus having a plurality of instructions stored thereon, the instructions being executable by a processing apparatus to operate a robot, the instructions configured to, when executed by the processing apparatus, cause the processing apparatus to:
operate in a first mode in which the robot executes actions in accordance with a training input and associates a first sensory context of a plurality of sensory contexts with a first behavior of a plurality of behaviors, the plurality of sensory contexts detected by at least one of a plurality of sensors of the robot;
configure a predictor to receive the plurality of sensory contexts and to determine one or more of the plurality of behaviors of the robot based at least in part on detection of the plurality of sensory contexts, wherein the predictor outputs a control signal based at least in part on priorities assigned to each of the plurality of behaviors for each of the plurality of sensory contexts; and
operate in a second mode in which the robot executes actions based at least in part on an association made while the robot operated in the first mode, wherein:
the robot detects: (1) the first sensory context with a first sensor of the plurality of sensors, and (2) a second sensory context of the plurality of sensory contexts with a second sensor; and
the robot performs the first behavior based at least in part on the control signal outputted by the predictor upon receiving the first sensory context and the second sensory context.
18. The storage apparatus of claim 17 , wherein the actions executed in the second mode comprise path navigation.
19. The storage apparatus of claim 17 , wherein the first sensor comprises a camera.
20. The storage apparatus of claim 17 , wherein the second sensor comprises a LiDAR.