Motion prediction for autonomous devices

1. A computer-implemented method of motion prediction, the computer-implemented method comprising:

accessing state data comprising information associated with one or more respective locations and characteristics of one or more objects over a plurality of time intervals, wherein the one or more objects comprise a two-axle vehicle;

determining one or more trajectories of the one or more objects based at least in part on the state data and a machine-learned tracking and kinematics model comprising a neural network trained by evaluation of a loss function to determine a distance between a front axle and a rear axle of the two-axle vehicle based on physical dimensions of the two-axle vehicle provided within the state data, wherein the one or more trajectories comprise one or more predicted locations of the one or more objects at one or more subsequent time intervals, and wherein the one or more predicted locations of the one or more objects are based at least in part on one or more physical constraints of the one or more objects, the one or more physical constraints comprising a turning angle of the one or more objects, wherein the turning angle of the two-axle vehicle is determined from the distance between the front axle and the rear axle of the two-axle vehicle and a velocity of the two-axle vehicle provided within the state data;

generating one or more outputs comprising one or more predicted object states based at least in part on the one or more predicted locations of the one or more objects at the one or more subsequent time intervals; and

controlling one or more vehicle systems associated with an autonomous vehicle based at least in part on the one or more predicted locations of the one or more objects.

2. The computer-implemented method of claim 1 , wherein:

the one or more objects comprise one or more moving objects; and

the determining the one or more trajectories of the one or more objects based at least in part on the state data and the machine-learned tracking and kinematics model comprises:

determining one or more predicted control inputs for the one or more moving objects based at least in part on the one or more trajectories of the one or more moving objects, wherein the one or more predicted control inputs comprise an input to one or more control devices of a moving object of the one or more moving objects.

3. The computer-implemented method of claim 2 , wherein the one or more predicted control inputs comprise a left turn of the one or more control devices or a right turn of the one or more control devices.

4. The computer-implemented method of claim 2 , further comprising

determining one or more control uncertainties respectively associated with the one or more predicted control inputs of the one or more objects at the one or more subsequent time intervals, wherein the one or more control uncertainties are respectively associated with a magnitude or degree of uncertainty for the one or more predicted control inputs of the one or more objects at the one or more subsequent time intervals.

5. The computer-implemented method of claim 1 , wherein the determining the one or more trajectories of the one or more objects based at least in part on the state data and the machine-learned tracking and kinematics model comprises:

determining one or more uncertainties respectively associated with the one or more predicted locations of the one or more objects, wherein the one or more uncertainties comprises a respective range distance around the one or more predicted locations.

6. The computer-implemented method of claim 1 , wherein the determining the one or more trajectories of the one or more objects based at least in part on the state data and the machine-learned tracking and kinematics model comprises:

determining the one or more physical constraints of the one or more objects based at least in part on the one or more respective locations and characteristics of the one or more objects.

7. The computer-implemented method of claim 6 , wherein the one or more physical constraints comprise at least one of a turning radius, a minimum stopping distance, a maximum acceleration, a maximum velocity, a maximum turn angle for each velocity of a plurality of velocities, a maximum jerk threshold, or a maximum velocity for each angle of a plurality of turn angles.

8. The computer-implemented method of claim 1 , wherein the state data is based at least in part on one or more outputs from one or more sensors comprising at least one of one or more light detection and ranging devices (LiDAR), one or more radar devices, one or more sonar devices, one or more thermal sensors, or one or more image sensors.

9. The computer-implemented method of claim 8 , wherein the one or more sensors are positioned to provide a bird's eye view of the one or more objects.

10. The computer-implemented method of claim 8 , wherein the state data comprises one or more rasterized RGB images of an environment detected by the one or more sensors.

11. The computer-implemented method of claim 1 , wherein the machine-learned tracking and kinematics model comprises a convolutional neural network including one or more tracking layers and one or more kinematic layers.

12. The computer-implemented method of claim 1 , wherein the one or more respective locations and characteristics comprise at least one of a plurality of physical dimensions, one or more shapes, an acceleration, a velocity, a heading, a heading rate of change, a latitude, a longitude, or an altitude.

13. The computer-implemented method of claim 12 , wherein the plurality of physical dimensions comprise a distance between a front wheel axle and a rear wheel axle of an object of the one or more objects.

14. The computer-implemented method of claim 1 , wherein the one or more objects comprise one or more vehicles with a fixed rear wheel axle and a moveable front wheel axle.

15. One or more tangible non-transitory computer-readable media storing computer-readable instructions that are executable by one or more processors to cause the one or more processors to perform operations, the operations comprising:

accessing state data comprising information associated with one or more respective locations and characteristics of one or more objects over a plurality of time intervals, wherein the one or more objects comprise a two-axle vehicle;

determining one or more trajectories of the one or more objects based at least in part on the state data and a machine-learned tracking and kinematics model comprising a neural network trained by evaluation of a loss function to determine a distance between a front axle and a rear axle of the two-axle vehicle based on physical dimensions of the two-axle vehicle provided within the state data, wherein the one or more trajectories comprise one or more predicted locations of the one or more objects at one or more subsequent time intervals, and wherein the one or more predicted locations of the one or more objects are based at least in part on one or more physical constraints of the one or more objects, the one or more physical constraints comprising a turning angle of the one or more objects, wherein the turning angle of the two-axle vehicle is determined from the distance between the front axle and the rear axle of the two-axle vehicle and a velocity of the two-axle vehicle provided within the state data;

generating one or more outputs comprising one or more predicted object states based at least in part on the one or more predicted locations of the one or more objects at the one or more subsequent time intervals; and

controlling one or more vehicle systems associated with an autonomous vehicle based at least in part on the one or more predicted locations of the one or more objects.

16. The one or more tangible non-transitory computer-readable media of claim 15 , wherein:

the one or more objects comprise one or more moving objects; and

the operations further comprise:

determining, based at least in part on the machine-learned tracking and kinematics model and the one or more trajectories of the one or more moving objects at the one or more subsequent time intervals, one or more predicted control inputs of the one or more moving objects at the one or more subsequent time intervals, wherein the one or more predicted control inputs of the one or more moving objects comprise one or more inputs to one or more control devices of a moving object of the one or more moving objects.

17. The one or more tangible non-transitory computer-readable media of claim 15 , the operations further comprising:

determining, based at least in part on the one or more predicted locations of the one or more objects, a travel path for a vehicle, wherein the travel path does not traverse the one or more predicted locations of the one or more objects at the time interval that the one or more objects are determined to be at the one or more predicted locations.

18. A computing device comprising:

one or more processors;

a memory comprising one or more computer-readable media, the memory storing computer-readable instructions that are executable by the one or more processors to cause the one or more processors to perform operations comprising:

accessing state data comprising information associated with one or more respective locations and characteristics of one or more objects over a plurality of time intervals, wherein the one or more objects comprise a two-axle vehicle;

determining one or more trajectories of the one or more objects based at least in part on the state data and a machine-learned tracking and kinematics model comprising a neural network trained by evaluation of a loss function to determine a distance between a front axle and a rear axle of the two-axle vehicle based on physical dimensions of the two-axle vehicle provided within the state data, wherein the one or more trajectories comprise one or more predicted locations of the one or more objects at one or more subsequent time intervals, and wherein the one or more predicted locations of the one or more objects are based at least in part on one or more physical constraints of the one or more objects, the one or more physical constraints comprising a turning angle of the one or more objects, wherein the turning angle of the two-axle vehicle is determined from the distance between the front axle and the rear axle of the two-axle vehicle and a velocity of the two-axle vehicle provided within the state data;

generating one or more outputs comprising one or more predicted object states based at least in part on the one or more predicted locations of the one or more objects at the one or more subsequent time intervals; and

controlling one or more vehicle systems associated with an autonomous vehicle based at least in part on the one or more predicted locations of the one or more objects.

19. The computing device of claim 18 , wherein the machine-learned tracking and kinematics model is based at least in part on training data comprising a plurality of sensor observations of a plurality of training objects traversing a plurality of paths at a plurality of velocities or a plurality of accelerations, and wherein each training object of the plurality of training objects is associated with a plurality of respective object labels comprising a distance between a front wheel axle and a rear wheel axle of each training object.

20. The computer-implemented method of claim 1 , wherein:

the machine-learned tracking and kinematics model is trained based at least in part on training data comprising a plurality of sensor observations of a plurality of training objects respectively associated with an object label including a distance between a front axle and a rear axle of the respective training object.