Vehicle control systems and methods using kinematically stabilized machine-learning model predicted controls
A method for controlling a vehicle based on a kinematically stable trajectory is described. The method includes receiving sensor data, from one or more sensor of an ego-vehicle configured to perceive an environment around the ego-vehicle; predicting, with a machine learning model configured to ingest the sensor data, an acceleration and a curvature defining a trajectory for the ego-vehicle; evaluating the acceleration and the curvature with a kinematics decoder configured to determine that the acceleration and the curvature are kinematically feasible for the ego-vehicle; generating, with the kinematics decoder, one or more outputs based on evaluation of the acceleration and the curvature, wherein the one or more outputs define the kinematically stable trajectory for controlling the ego-vehicle; and controlling the ego-vehicle based on the one or more outputs.
1 . A method for controlling an ego-vehicle based on a kinematically stable trajectory, the method comprising:
receiving sensor data, from one or more sensors of the ego-vehicle configured to perceive an environment around the ego-vehicle;
predicting, with a machine learning model configured to ingest the sensor data, an acceleration and a curvature defining a trajectory for the ego-vehicle;
evaluating the acceleration and the curvature with a kinematics decoder that incorporates knowledge of capabilities of the ego-vehicle to determine that the acceleration and the curvature are kinematically feasible for the ego-vehicle;
generating, with the kinematics decoder, one or more outputs based on evaluation of the acceleration and the curvature, wherein the one or more outputs define the kinematically stable trajectory for controlling the ego-vehicle; and
controlling the ego-vehicle based on the one or more outputs.
2 . The method of claim 1 , wherein the one or more outputs comprise a position, a yaw, and a velocity.
3 . The method of claim 1 , wherein the acceleration and the curvature defining the trajectory for the ego-vehicle comprises a predicted series of accelerations and curvatures and the kinematics decoder is configured to evaluate the predicted series of accelerations and curvatures to generate the one or more outputs, wherein the one or more outputs define the kinematically stable trajectory for the ego-vehicle.
4 . The method of claim 1 , wherein the kinematics decoder invokes a unicycle model for evaluating the acceleration and the curvature.
5 . The method of claim 1 , wherein evaluating the acceleration and the curvature with the kinematics decoder comprises applying one or more constraints defined in the kinematics decoder, the one or more constraints comprise at least one of a threshold velocity, a threshold acceleration, or a threshold steering angle corresponding to the ego-vehicle.
6 . The method of claim 1 , further comprising:
determining, with the kinematics decoder, one or more constraints based on sensor data and a history of trajectories of the ego-vehicle, and
adjusting, with the kinematics decoder, at least one of the acceleration or the curvature based on the one or more constraints.
7 . The method of claim 1 , wherein the sensor data comprises image or video data and distance and ranging data from at least one of a camera system, a LiDAR system, or a radar system.
8 . The method of claim 1 , wherein predicting, with the machine learning model, the acceleration and the curvature defining the trajectory for the ego-vehicle comprises:
determining a future trajectory of one or more other objects in the environment around the ego-vehicle based on the sensor data; and
determining the acceleration and the curvature defining the trajectory for the ego-vehicle based on the future trajectory of the one or more other objects.
9 . An apparatus for controlling an ego-vehicle based on a kinematically stable trajectory, comprising: one or more memories comprising processor-executable instructions; and one or more processors configured to execute the processor-executable instructions and cause the apparatus to:
receive sensor data, from one or more sensors of the ego-vehicle configured to perceive an environment around the ego-vehicle;
predict, with a machine learning model configured to ingest the sensor data, an acceleration and a curvature defining a trajectory for the ego-vehicle;
evaluate the acceleration and the curvature with a kinematics decoder that incorporates knowledge of capabilities of the ego-vehicle to determine that the acceleration and the curvature are kinematically feasible for the ego-vehicle;
generate, with the kinematics decoder, one or more outputs based on evaluation of the acceleration and the curvature, wherein the one or more outputs define the kinematically stable trajectory for controlling the ego-vehicle; and
control the ego-vehicle based on the one or more outputs.
10 . The apparatus of claim 9 , wherein the one or more outputs comprise a position, a yaw, and a velocity.
11 . The apparatus of claim 9 , wherein the acceleration and the curvature defining the trajectory for the ego-vehicle comprises a predicted series of accelerations and curvatures and the kinematics decoder is configured to evaluate the predicted series of accelerations and curvatures to generate the one or more outputs, wherein the one or more outputs define the kinematically stable trajectory for the ego-vehicle.
12 . The apparatus of claim 9 , wherein the kinematics decoder invokes a unicycle model for evaluating the acceleration and the curvature.
13 . The apparatus of claim 9 , wherein to evaluate the acceleration and the curvature with the kinematics decoder, the one or more processors are configured to apply one or more constraints defined in the kinematics decoder, the one or more constraints comprise at least one of a threshold velocity, a threshold acceleration, or a threshold steering angle corresponding to the ego-vehicle.
14 . The apparatus of claim 9 , wherein the one or more processors are configured to execute the processor-executable instructions and further cause the apparatus to:
determine, with the kinematics decoder, one or more constraints based on sensor data and a history of trajectories of the ego-vehicle, and
adjust, with the kinematics decoder, at least one of the acceleration or the curvature based on the one or more constraints.
15 . The apparatus of claim 9 , wherein the sensor data comprises image or video data and distance and ranging data from at least one of a camera system, a LiDAR system, or a radar system.
16 . The apparatus of claim 9 , wherein to predict, with the machine learning model, the acceleration and the curvature defining the trajectory for the ego-vehicle, the one or more processors are configured to:
determine a future trajectory of one or more other objects in the environment around the ego-vehicle based on the sensor data; and
determine the acceleration and the curvature defining the trajectory for the ego-vehicle based on the future trajectory of the one or more other objects.
17 . A vehicle comprising:
one or more sensors configured to collect sensor data of an environment around the vehicle; and
a computing device comprising one or more memories comprising processor-executable instructions; and one or more processors configured to execute the processor-executable instructions causing the computing device to:
predict, with a machine learning model configured to ingest the sensor data, an acceleration and a curvature defining a trajectory for the vehicle;
evaluate the acceleration and the curvature with a kinematics decoder that incorporates knowledge of capabilities of the vehicle to determine that the acceleration and the curvature are kinematically feasible for the vehicle;
generate, with the kinematics decoder, one or more outputs based on evaluation of the acceleration and the curvature, wherein the one or more outputs define a kinematically stable trajectory for controlling the vehicle; and
control the vehicle based on the one or more outputs.
18 . The vehicle of claim 17 , wherein the acceleration and the curvature defining the trajectory for the vehicle comprises a predicted series of accelerations and curvatures and the kinematics decoder is configured to evaluate the predicted series of accelerations and curvatures to generate the one or more outputs, wherein the one or more outputs define the kinematically stable trajectory for the vehicle.
19 . The vehicle of claim 17 , wherein the one or more processors are configured to execute the processor-executable instructions and further cause the computing device to:
determine, with the kinematics decoder, one or more constraints based on sensor data and a history of trajectories of the vehicle, and
adjust, with the kinematics decoder, at least one of the acceleration or the curvature based on the one or more constraints.
20 . The vehicle of claim 17 , wherein to predict, with the machine learning model, the acceleration and the curvature defining the trajectory for the vehicle, the one or more processors are configured to:
determine a future trajectory of one or more other objects in the environment around the vehicle based on the sensor data; and
determine the acceleration and the curvature defining the trajectory for the vehicle based on the future trajectory of the one or more other objects.