Shared decision-making with cognitive priors
Systems, methods, and other embodiments described herein relate to integrating human decision-making into a model-based system. In one embodiment, a method includes acquiring sensor data, including driver data about a driver of a vehicle and driving data about the vehicle and a surrounding environment of the vehicle. The method includes encoding, using a world encoder, the sensor data into a latent representation. The method includes determining human decision-making characteristics according to the latent representation. The method includes generating a control signal for providing shared control of the vehicle according to the human decision-making characteristics and the latent representation.
1 . A cognitive system, comprising:
one or more processors;
a memory communicably coupled to the one or more processors and storing instructions that, when executed by the one or more processors, cause the one or more processors to:
acquire sensor data, including driver data about a driver of a vehicle and driving data about the vehicle and a surrounding environment of the vehicle;
encode, using a world encoder, the sensor data into a latent representation, the latent representation being encoded features of the sensor data within a defined latent space;
determine human decision-making characteristics that define at least an intent of the driver and latencies of the driver in making a decision according to the latent representation; and
generate a control signal that provides shared control of the vehicle according to the human decision-making characteristics and the latent representation, the shared control including blending driver control inputs with automated control inputs.
2 . The cognitive system of claim 1 , wherein the instructions to encode the sensor data into the latent representation include instructions to encode a latent state of the driver to integrate human decision-making within the latent representation.
3 . The cognitive system of claim 1 , wherein the instructions to determine the human decision-making characteristics include instructions to apply a decision model to predict human decision-making and latency in making a decision from the latent representation.
4 . The cognitive system of claim 3 , wherein the decision model implements one of nonlinear opinion dynamics and drift-diffusion modeling to infer what the driver is considering, limitations of the driver in relation to the human decision-making, and to account for latencies in making the decision, and
wherein the instructions to determine the human decision-making characteristics include instructions to infer, using a dynamics model, a subsequent representation of the latent representation at a subsequent time step to further account for future considerations of the driver.
5 . The cognitive system of claim 1 , wherein the instructions to determine the human decision-making characteristics includes applying a decision model to generate outputs indicating the intent of the driver, preferences for decisions, and preferences for a mode of interaction according to the latent representation.
6 . The cognitive system of claim 1 , wherein generating the control signal includes applying a policy model that determines how to blend driver control inputs with automated control inputs for the shared control according to the human decision-making characteristics and the latent representation, and
wherein the policy model is a semi-Markov model (SMM).
7 . The cognitive system of claim 1 , further comprising:
training at least the world encoder and a policy model using a reward network that predicts rewards according to the latent representation, wherein training includes training a decision model using labeled data.
8 . The cognitive system of claim 1 , wherein the cognitive system is integrated within the vehicle that operates at least semi-autonomously.
9 . A non-transitory computer-readable medium including instructions that, when executed by one or more processors, cause the one or more processors to:
acquire sensor data, including driver data about a driver of a vehicle and driving data about the vehicle and a surrounding environment of the vehicle;
encode, using a world encoder, the sensor data into a latent representation, the latent representation being encoded features of the sensor data within a defined latent space;
determine human decision-making characteristics that define at least an intent of the driver and latencies of the driver in making a decision according to the latent representation; and
generate a control signal for providing shared control of the vehicle according to the human decision-making characteristics and the latent representation, the shared control including blending driver control inputs with automated control inputs.
10 . The non-transitory computer-readable medium of claim 9 , wherein the instructions to encode the sensor data into the latent representation include instructions to encode a latent state of the driver to integrate human decision-making within the latent representation.
11 . The non-transitory computer-readable medium of claim 9 , wherein the instructions to determine the human decision-making characteristics include instructions to apply a decision model to predict human decision-making and latency in making a decision from the latent representation.
12 . The non-transitory computer-readable medium of claim 11 , wherein the decision model implements one of nonlinear opinion dynamics and drift-diffusion modeling to infer what the driver is considering, limitations of the driver in relation to the human decision-making, and to account for latencies in making the decision, and
wherein the instructions to determine the human decision-making characteristics include instructions to infer, using a dynamics model, a subsequent representation of the latent representation at a subsequent time step to further account for future considerations of the driver.
13 . The non-transitory computer-readable medium of claim 9 , wherein the instructions to determine the human decision-making characteristics includes applying a decision model to generate outputs indicating intent of the driver, preferences for decisions, and preferences for a mode of interaction according to the latent representation.
14 . A method, comprising:
acquiring sensor data, including driver data about a driver of a vehicle and driving data about the vehicle and a surrounding environment of the vehicle;
encoding, using a world encoder, the sensor data into a latent representation, the latent representation being encoded features of the sensor data within a defined latent space;
determining human decision-making characteristics that define at least an intent of the driver and latencies of the driver in making a decision according to the latent representation; and
generating a control signal that provides shared control of the vehicle according to the human decision-making characteristics and the latent representation, the shared control including blending driver control inputs with automated control inputs.
15 . The method of claim 14 , wherein encoding the sensor data into the latent representation includes encoding a latent state of the driver to integrate human decision-making within the latent representation.
16 . The method of claim 14 , wherein determining the human decision-making characteristics includes applying a decision model to predict human decision-making and latency in making a decision from the latent representation.
17 . The method of claim 16 , wherein the decision model implements one of nonlinear opinion dynamics and drift-diffusion modeling to infer what the driver is considering, limitations of the driver in relation to the human decision-making, and to account for latencies in making the decision, and
wherein determining the human decision-making characteristics includes inferring, using a dynamics model, a subsequent representation of the latent representation at a subsequent time step to further account for future considerations of the driver.
18 . The method of claim 14 , wherein determining the human decision-making characteristics includes applying a decision model to generate outputs indicating the intent of the driver, preferences for decisions, and preferences for a mode of interaction according to the latent representation.
19 . The method of claim 14 , wherein generating the control signal includes applying a policy model that determines how to blend driver control inputs with automated control inputs for the shared control according to the human decision-making characteristics and the latent representation,
wherein the policy model is a semi-Markov model (SMM).
20 . The method of claim 14 , further comprising:
training at least the world encoder and a policy model using a reward network that predicts rewards according to the latent representation, wherein training includes training a decision model using labeled data.