Systems and methods for predicting the trajectory of a moving object
Systems and methods for predicting a trajectory of a moving object are disclosed herein. One embodiment downloads, to a robot, a probabilistic hybrid discrete-continuous automaton (PHA) model learned as a deep neural network; uses the deep neural network to infer a sequence of high-level discrete modes and a set of associated low-level samples, wherein the high-level discrete modes correspond to candidate maneuvers for the moving object and the low-level samples are candidate trajectories; uses the sequence of high-level discrete modes and the set of associated low-level samples, via a learned proposal distribution in the deep neural network, to adaptively sample the sequence of high-level discrete modes to produce a reduced set of low-level samples; applies a sample selection technique to the reduced set of low-level samples to select a predicted trajectory for the moving object; and controls operation of the robot based, at least in part, on the predicted trajectory.
1. A system for predicting a trajectory of a moving object, the system comprising:
a processor; and
a memory storing instructions that, when executed by the processor, cause the processor to:
download, to a robot, a probabilistic hybrid discrete-continuous automaton (PHA) model learned as a deep neural network, wherein the learned PHA model models a moving object in an environment of the robot;
use the deep neural network to infer a sequence of high-level discrete modes and a set of associated low-level samples, wherein the high-level discrete modes correspond to candidate maneuvers for the moving object and the low-level samples are candidate trajectories for the moving object;
use the sequence of high-level discrete modes and the set of associated low-level samples, via a learned proposal distribution in the deep neural network, to adaptively sample the sequence of high-level discrete modes to produce a reduced set of low-level samples; and
apply a sample selection technique to the reduced set of low-level samples to select a predicted trajectory for the moving object; and
control operation of the robot based, at least in part, on the predicted trajectory.
2. The system of claim 1 , wherein:
the deep neural network includes an encoder Long Short-Term Memory (LSTM) network that encodes context information concerning the environment of the robot and a decoder LSTM network that generates a sequence of hybrid states in accordance with the learned PHA model; and
the decoder LSTM network includes a transition function, a dynamics function, and the learned proposal distribution.
3. The system of claim 2 , wherein the context information includes one or more of environment-sensor data, map data, observation data pertaining to the moving object, trajectory data associated with at least one of the robot and the moving object, traffic Signal Phase and Timing (SPaT) data, and information received via vehicle-to-vehicle (V2V) communication.
4. The system of claim 1 , wherein the sample selection technique includes a farthest point sampling algorithm.
5. The system of claim 1 , wherein the robot is a vehicle and the moving object is a road agent external to the vehicle.
6. The system of claim 5 , wherein the vehicle is an autonomous vehicle.
7. The system of claim 1 , wherein the robot is a vehicle and the moving object is the vehicle.
8. The system of claim 7 , wherein the vehicle is an autonomous vehicle and a planner of the autonomous vehicle uses the predicted trajectory in planning a path for the autonomous vehicle.
9. The system of claim 1 , wherein the robot is one of an indoor robot, a service robot, and a delivery robot.
10. A non-transitory computer-readable medium for predicting a trajectory of a moving object and storing instructions that, when executed by a processor, cause the processor to:
download, to a robot, a probabilistic hybrid discrete-continuous automaton (PHA) model learned as a deep neural network, wherein the learned PHA model models a moving object in an environment of the robot;
use the deep neural network to infer a sequence of high-level discrete modes and a set of associated low-level samples, wherein the high-level discrete modes correspond to candidate maneuvers for the moving object and the low-level samples are candidate trajectories for the moving object;
use the sequence of high-level discrete modes and the set of associated low-level samples, via a learned proposal distribution in the deep neural network, to adaptively sample the sequence of high-level discrete modes to produce a reduced set of low-level samples;
apply a sample selection technique to the reduced set of low-level samples to select a predicted trajectory for the moving object; and
control operation of the robot based, at least in part, on the predicted trajectory.
11. The non-transitory computer-readable medium of claim 10 , wherein:
the deep neural network includes an encoder Long Short-Term Memory (LSTM) network that encodes context information concerning the environment of the robot and a decoder LSTM network that generates a sequence of hybrid states in accordance with the learned PHA model; and
the decoder LSTM network includes a transition function, a dynamics function, and the learned proposal distribution.
12. The non-transitory computer-readable medium of claim 10 , wherein the sample selection technique includes a farthest point sampling algorithm.
13. A method, comprising:
downloading, to a robot, a probabilistic hybrid discrete-continuous automaton (PHA) model learned as a deep neural network, wherein the learned PHA model models a moving object in an environment of the robot;
using the deep neural network to infer a sequence of high-level discrete modes and a set of associated low-level samples, wherein the high-level discrete modes correspond to candidate maneuvers for the moving object and the low-level samples are candidate trajectories for the moving object;
using the sequence of high-level discrete modes and the set of associated low-level samples, via a learned proposal distribution in the deep neural network, to adaptively sample the sequence of high-level discrete modes to produce a reduced set of low-level samples;
applying a sample selection technique to the reduced set of low-level samples to select a predicted trajectory for the moving object; and
controlling operation of the robot based, at least in part, on the predicted trajectory.
14. The method of claim 13 , wherein:
the deep neural network includes an encoder Long Short-Term Memory (LSTM) network that encodes context information concerning the environment of the robot and a decoder LSTM network that generates a sequence of hybrid states in accordance with the learned PHA model; and
the decoder LSTM network includes a transition function, a dynamics function, and the learned proposal distribution.
15. The method of claim 13 , wherein the sample selection technique includes a farthest point sampling algorithm.
16. The method of claim 13 , wherein the robot is a vehicle and the moving object is a road agent external to the vehicle.
17. The method of claim 16 , wherein the vehicle is an autonomous vehicle.
18. The method of claim 13 , wherein the robot is a vehicle and the moving object is the vehicle.
19. The method of claim 18 , wherein the vehicle is an autonomous vehicle and a planner of the autonomous vehicle uses the predicted trajectory in planning a path for the autonomous vehicle.
20. The method of claim 13 , wherein the robot is one of an indoor robot, a service robot, and a delivery robot.