Automatic navigation using deep reinforcement learning

1. A method comprising:

identifying a state of an autonomous vehicle within a simulated environment, wherein the simulated environment comprises a plurality of operating parameters for training a reinforcement learning framework to calculate driving maneuvers for the autonomous vehicle;

calculating, with an actor neural network, a driving maneuver for navigating the autonomous vehicle from an initial location to a target destination;

determining, with a critic neural network, whether the driving maneuver was beneficial for accurately maneuvering the autonomous vehicle to the target destination;

in response to determining the driving maneuver was beneficial, causing the critic neural network to reward the actor neural network during a training phase for the reinforcement learning framework; and

in response to determining the driving maneuver was not beneficial, causing the critic neural network to penalize the actor neural network during the training phase for the reinforcement learning framework;

wherein the critic neural network rewards the actor neural network by providing an error signal calculated with an exploration-exploitation tradeoff model.

2. The method of claim 1 , wherein each of the actor neural network and the critic neural network are installed onboard the autonomous vehicle during the training phase for the reinforcement learning framework.

3. The method of claim 1 , wherein the autonomous vehicle comprises at least one sensor selected from a group consisting of a camera sensor, a lidar sensor, a radar sensor, a GPS sensor, and an ultrasound sensor.

4. The method of claim 1 , further comprising determining a state of the autonomous vehicle within the simulated environment, wherein the state comprises one or more of a location or an orientation of the autonomous vehicle.

5. The method of claim 1 , wherein the critic neural network penalizes the actor neural network by providing an error signal calculated with an exploration-exploitation tradeoff model.

6. The method of claim 1 , further comprising storing one or more of a state of the autonomous vehicle, an action taken at the state of the autonomous vehicle, or a reward and a penalty corresponding to the action in a replay buffer.

7. The method of claim 6 , further comprising sampling the replay buffer to train the actor neural network.

8. The method of claim 7 , further comprising iteratively navigating the autonomous vehicle from the initial location to the target destination in accordance with the training.

9. The method of claim 1 , wherein calculating the driving maneuver further comprises:

calculating a plurality of driving maneuvers for a plurality of autonomous vehicles from the initial location to the target destination, wherein the plurality of autonomous vehicles comprises the autonomous vehicle; and

communicating information from a neural network corresponding to each of the plurality of autonomous vehicles to a central master actor.

10. A system comprising:

an autonomous vehicle comprising a sensor;

one or more processors configurable to execute instructions stored in non-transitory computer readable memory, the instructions comprising:

identifying a state of an autonomous vehicle within a simulated environment, wherein the simulated environment comprises a plurality of operating parameters for training a reinforcement learning framework to calculate driving maneuvers for the autonomous vehicle;

calculating, with an actor neural network, a driving maneuver for navigating the autonomous vehicle from an initial location to a target destination;

determining, with a critic neural network, whether the driving maneuver was beneficial for accurately maneuvering the autonomous vehicle to the target destination;

in response to determining the driving maneuver was beneficial, causing the critic neural network to reward the actor neural network during a training phase for the reinforcement learning framework; and

in response to determining the driving maneuver was not beneficial, causing the critic neural network to penalize the actor neural network during the training phase for the reinforcement learning framework;

wherein the critic neural network rewards the actor neural network by providing an error signal calculated with an exploration-exploitation tradeoff model.

11. The system of claim 10 , wherein each of the actor neural network and the critic neural network are installed onboard the autonomous vehicle during the training phase for the reinforcement learning framework.

12. The system of claim 10 , wherein the sensor of the autonomous vehicle comprises one or more of a camera sensor, a lidar sensor, a radar sensor, a GPS sensor, and an ultrasound sensor.

13. The system of claim 10 , wherein the instructions further comprise determining a state of the autonomous vehicle within the simulated environment, wherein the state of the autonomous vehicle comprises one or more of a location or an orientation of the autonomous vehicle.

14. The system of claim 10 , wherein the critic neural network rewards the actor neural network by providing an error signal calculated using an exploration-exploitation tradeoff model.

15. The system of claim 10 , wherein the instructions further comprise storing one or more of a state of the autonomous vehicle, an action taken at the state of the autonomous vehicle, or a reward and a penalty corresponding to the action in a replay buffer.

16. The system of claim 10 , wherein the instructions further comprise:

calculating a new plurality of driving maneuvers for a plurality of autonomous vehicles from the initial location to the target destination, wherein the plurality of autonomous vehicles comprises the autonomous vehicle; and

communicating information from a neural network corresponding to each of the plurality of autonomous vehicles to a central master actor.

17. Non-transitory computer readable storage medium storing instructions for execution by one or more processors, the instructions comprising:

identifying a state of an autonomous vehicle within a simulated environment, wherein the simulated environment comprises a plurality of operating parameters for training a reinforcement learning framework to calculate driving maneuvers for the autonomous vehicle;

calculating, with an actor neural network, a driving maneuver for navigating the autonomous vehicle from an initial location to a target destination;

determining, with a critic neural network, whether the driving maneuver was beneficial for accurately maneuvering the autonomous vehicle to the target destination;

in response to determining the driving maneuver was beneficial, causing the critic neural network to reward the actor during a training phase for the reinforcement learning framework; and

in response to determining the driving maneuver was not beneficial, causing the critic neural network to penalize the actor neural network during the training phase for the reinforcement learning framework;

wherein the critic neural network rewards the actor neural network by providing an error signal calculated with an exploration-exploitation tradeoff model.

18. The non-transitory computer readable storage medium of claim 17 , wherein the instructions further comprise:

calculating a new plurality of driving maneuvers for a plurality of autonomous vehicles from the initial location to the target destination, wherein the plurality of autonomous vehicles comprises the autonomous vehicle; and

communicating information from a neural network corresponding to each of the plurality of autonomous vehicles to a central master actor.