Context aware optimization of planner for autonomous driving
Systems and methods for optimizing a planning model for autonomous driving. Image data is generated from a vehicle camera. A perception model detects agents in the environment based on the image data. A planner model is executed to generate a predicted trajectory of the vehicle based upon configuration parameters associated with characteristics of the vehicle. Information regarding a context associated with the detected agents is received from the perception model. An optimizer is configured to select a subset of the configuration parameters to be optimized based on the context, select an objective function based on the context, adjust the selected subset of configuration parameters based on a current value of the objective function, generate a new planned trajectory, derive a value of the objective function, and determine optimal configuration parameters that minimize the objective function.
1 . A method for optimizing a planner model for autonomous driving, the method comprising:
receiving image data generated from a camera mounted to a vehicle, wherein the image data includes agents in an environment outside the vehicle;
via a pre-existing perception model, detecting the agents in the environment based on the image data;
executing a planner model to generate a predicted trajectory of the vehicle, wherein the generated predicted trajectory of the vehicle is dependent upon a set of configuration parameters associated with characteristics of the vehicle;
receiving, from the perception model, information regarding a context associated with the detected agents in the environment, wherein the context comprises both (i) predicted motion vectors for each detected agent in the environment and (ii) categorical road-type label regarding a type of road; and
executing a black-box optimizer configured to optimize performance of the planner model based on the context, wherein execution of the optimizer:
(1) selects a subset of the configuration parameters to be optimized based on the context,
(2) selects an objective function based on the context,
(3) adjusts the selected subset of the configuration parameters based on a current value of the objective function,
(4) generates a new planned trajectory of the vehicle based on the adjusted configuration parameters,
(5) derives a value of the objective function based on the new planned trajectory, and
(6) repeats (3)-(5) to determine optimal configuration parameters that minimize the objective function.
2 . The method of claim 1 , wherein the black-box optimizer is a Bayesian optimizer.
3 . The method of claim 1 , wherein the subset of configuration parameters is related to a vehicle controller system;
wherein the selected objective function includes an average displacement error calculated by, for each subset of configuration parameters evaluated during the optimization, generating a planned trajectory of the vehicle and determining the average displacement error between the planned trajectory and a ground-truth trajectory of the vehicle, wherein the ground-truth trajectory is obtained from previously recorded sensor data prior to execution of the optimizer.
4 . The method of claim 3 , wherein execution of the optimizer:
samples the subset of the configuration parameters,
generates the new planned trajectory based on the adjustment of the selected subset of the configuration parameters, and
calculates the average displacement error.
5 . The method of claim 1 , wherein the planning model includes a planning algorithm, and the subset of configuration parameters are related to the planning algorithm;
wherein the selected objective function includes a loss function of the planning algorithm.
6 . The method of claim 5 , wherein execution of the optimizer:
samples the subset of the configuration parameters,
generates the new planned trajectory based on the adjustment of the selected subset of the configuration parameters, and
calculates the loss function of the planning algorithm.
7 . The method of claim 1 , wherein the detecting of the agents in the environment is further based on a pre-existing localization model.
8 . The method of claim 1 , further comprising:
receiving sensor data from sensors mounted to the vehicle, wherein the sensors are not cameras,
wherein the detecting of the agents in the environment is further based on the sensor data.
9 . The method of claim 1 , wherein the context associated with the detected agents is associated with movement of the detected agents and a characteristic of a road that the vehicle is traveling on.
10 . The method of claim 1 , wherein the characteristic of the road that the vehicle is traveling on includes whether the road is one of a highway, urban road, or roundabout.
11 . A system for optimizing a planner model for autonomous driving, the system comprising:
a processor; and
memory including instructions that, when executing by the processor, cause the processor to perform the following steps:
receiving image data generated from a camera mounted to a vehicle, wherein the image data includes agents in an environment outside the vehicle;
via a pre-existing perception model, detecting the agents in the environment based on the image data;
executing a planner model to generate a predicted trajectory of the vehicle, wherein the generated predicted trajectory of the vehicle is dependent upon a set of configuration parameters associated with characteristics of the vehicle;
receiving, from the perception model, information regarding a context associated with the detected agents in the environment, wherein the context comprises both (i) predicted motion vectors for each detected agent in the environment and (ii) categorical road-type label regarding a type of road; and
executing a black-box optimizer configured to optimize performance of the planner model based on the context, wherein execution of the optimizer:
(1) selects a subset of the configuration parameters to be optimized based on the context,
(2) selects an objective function based on the context,
(3) adjusts the selected subset of the configuration parameters based on a current value of the objective function,
(4) generates a new planned trajectory of the vehicle based on the adjusted configuration parameters,
(5) derives a value of the objective function based on the new planned trajectory, and
(6) repeats (3)-(5) to determine optimal configuration parameters that minimize the objective function.
12 . The system of claim 11 , wherein the black-box optimizer is a Bayesian optimizer.
13 . The system of claim 11 , wherein the subset of configuration parameters is related to a vehicle controller system; and
wherein the selected objective function includes an average displacement error calculated by, for each set of configuration parameters evaluated during the optimization, generating a planned trajectory of the vehicle and determining the average displacement error between the planned trajectory and a ground-truth trajectory of the vehicle, wherein the ground-truth trajectory is obtained from previously recorded sensor data prior to execution of the optimizer.
14 . The system of claim 13 , wherein execution of the optimizer:
samples the subset of the configuration parameters,
generates the new planned trajectory based on the adjustment of the selected subset of the configuration parameters, and
calculates the average displacement error.
15 . The system of claim 11 , wherein the planning model includes a planning algorithm, and the subset of configuration parameters are related to the planning algorithm;
wherein the selected objective function includes a loss function of the planning algorithm.
16 . The system of claim 15 , wherein execution of the optimizer:
samples the subset of the configuration parameters,
generates the new planned trajectory based on the adjustment of the selected subset of the configuration parameters, and
calculates the loss function of the planning algorithm.
17 . The system of claim 11 , wherein the detecting of the agents in the environment is further based on a pre-existing localization model.
18 . The system of claim 11 , wherein the instructions are further configured to, when executed by the processor, cause the processor to perform the following:
receiving sensor data from sensors mounted to the vehicle, wherein the sensors are not cameras,
wherein the detecting of the agents in the environment is further based on the sensor data.
19 . The system of claim 11 , wherein the context associated with the detected agents is associated with movement of the detected agents and a characteristic of a road that the vehicle is traveling on.
20 . A non-transitory computer-readable storage medium comprising instructions that, when executed by a processor, cause the processor to perform:
receive image data generated from a camera mounted to a vehicle, wherein the image data includes agents in an environment outside the vehicle;
via a pre-existing perception model, detect the agents in the environment based on the image data;
execute a planner model to generate a predicted trajectory of the vehicle, wherein the generated predicted trajectory of the vehicle is dependent upon a set of configuration parameters associated with characteristics of the vehicle;
receive, from the perception model, information regarding a context associated with the detected agents in the environment, wherein the context comprises both (i) predicted motion vectors for each detected agent in the environment and (ii) categorical road-type label regarding a type of road; and
execute a Bayesian optimizer configured to optimize performance of the planner model based on the context, wherein execution of the optimizer:
(1) selects a subset of the configuration parameters to be optimized based on the context,
(2) selects an objective function based on the context,
(3) adjusts the selected subset of the configuration parameters based on a current value of the objective function,
(4) generates a new planned trajectory of the vehicle based on the adjusted configuration parameters,
(5) derives a value of the objective function based on the new planned trajectory, and
(6) repeats (3)-(5) to determine optimal configuration parameters that minimize the objective function.