Systems and methods for risk-sensitive sequential action control for robotic devices

1. A control system for improving controls in a device according to risk, comprising:

one or more processors;

a memory communicably coupled to the one or more processors and storing:

a detection module including instructions that, when executed by the one or more processors, cause the one or more processors to, in response to receiving sensor data about a surrounding environment of the device, identify objects from the sensor data that are present in the surrounding environment; and

a control module including instructions that, when executed by the one or more processors, cause the one or more processors to generate a control sequence for controlling the device according to a risk-sensitivity parameter to navigate toward a destination while considering risk associated with encountering the objects defined by the risk-sensitivity parameter, the risk-sensitivity parameter being selectable along a continuum to define an extent of variability that is acceptable in a collision cost in addition to a mean for controlling the device, and

control the device according to the control sequence.

2. The control system of claim 1 , wherein the control module includes instructions to generate the control sequence including instructions to perform mode insertion gradient optimization of a risk objective embodied by the risk parameter using dynamics of the device and sampled costs associated with trajectories of the objects, and

wherein the risk-sensitivity parameter accounting for a distribution of the collision cost to adapt generating the control sequence beyond a discrete optimization.

3. The control system of claim 1 , wherein the control module includes instructions to generate the control sequence by performing mode insertion gradient optimization including simulating dynamics of the device forward in time out to a prediction horizon.

4. The control system of claim 1 , wherein the control module includes instructions to generate the control sequence by performing mode insertion gradient optimization including instructions to estimate the control sequence to refine a nominal sequence according to the risk parameter and according to trajectories of the objects in order to avoid collisions with the objects.

5. The control system of claim 1 , wherein the control module includes instructions to control the device including instructions to apply one or more controls specified by the control sequence to control movements of the device, and

wherein the control module includes instructions to generate the control sequence iteratively to refine the control sequence according to a quadratic optimization in stochastic sequential action control.

6. The control system of claim 1 , wherein the detection module includes instructions to identify the objects including instructions to predict trajectories of the objects according to a generative behavior prediction model out to a prediction time horizon, and

wherein the control module includes instructions to generate the control sequence according to a risk-sensitive sequential action control algorithm that is a sampling-based stochastic model predictive control approach that is risk aware.

7. The control system of claim 1 , wherein the detection module includes instructions to identify the objects including instructions to determine an initial state of the device, including dynamics and a current location, a reference trajectory indicating a desired path of the device, and a nominal control schedule for controlling the device over the reference trajectory.

8. The control system of claim 1 , wherein the device is an autonomous vehicle.

9. A non-transitory computer-readable medium for improving controls in a device according to risk and including instructions that when executed by one or more processors cause the one or more processors to:

in response to receiving sensor data about a surrounding environment of the device, identify objects from the sensor data that are present in the surrounding environment;

generate a control sequence for controlling the device according to a risk-sensitivity parameter to navigate toward a destination while considering risk associated with encountering the objects defined by the risk-sensitivity parameter, the risk-sensitivity parameter being selectable along a continuum to define an extent of variability that is acceptable in a collision cost in addition to a mean for controlling the device; and

control the device according to the control sequence.

10. The non-transitory computer-readable medium of claim 9 , wherein the instructions to generate the control sequence include instructions to perform mode insertion gradient optimization of a risk objective embodied by the risk parameter using dynamics of the device and sampled costs associated with trajectories of the objects, and

wherein the risk-sensitivity parameter accounting for a distribution of the collision cost to adapt generating the control sequence beyond a discrete optimization.

11. The non-transitory computer-readable medium of claim 9 , wherein the instructions to control the device include instructions to apply one or more controls specified by the control sequence to control movements of the device, and

wherein the instructions to generate the control sequence execute iteratively to refine the control sequence according to a quadratic optimization in stochastic sequential action control.

12. The non-transitory computer-readable medium of claim 9 , wherein the instructions to identify the objects include instructions to predict trajectories of the objects according to a generative behavior prediction model out to a prediction time horizon, and

wherein the instructions to generate the control sequence apply a risk-sensitive sequential action control algorithm that is a sampling-based stochastic model predictive control approach that is risk aware.

13. The non-transitory computer-readable medium of claim 9 , wherein the instructions to identify the objects include instructions to determine an initial state of the device, including dynamics and a current location, a reference trajectory indicating a desired path of the device, and a nominal control schedule for controlling the device over the reference trajectory.

14. A method of improving controls in a device according to risk, comprising:

in response to receiving sensor data about a surrounding environment of the device, identifying objects from the sensor data that are present in the surrounding environment;

generating a control sequence for controlling the device according to a risk-sensitivity parameter to navigate toward a destination while considering risk associated with encountering the objects defined by the risk-sensitivity parameter, the risk-sensitivity parameter being selectable along a continuum to define an extent of variability that is acceptable in a collision cost in addition to a mean for controlling the device; and

controlling the device according to the control sequence.

15. The method of claim 14 , wherein generating the control sequence includes performing mode insertion gradient optimization of a risk objective embodied by the risk parameter using dynamics of the device and sampled costs associated with trajectories of the objects, and

wherein the risk-sensitivity parameter accounting for a distribution of the collision cost to adapt generating the control sequence beyond a discrete optimization.

16. The method of claim 14 , wherein generating the control sequence by performing mode insertion gradient optimization includes simulating dynamics of the device forward in time out to a prediction horizon.

17. The method of claim 14 , wherein generating the control sequence by performing mode insertion gradient optimization includes estimating the control sequence to refine a nominal sequence according to the risk parameter and according to trajectories of the objects in order to avoid collisions with the objects.

18. The method of claim 14 , wherein controlling the device includes applying one or more controls specified by the control sequence to control movements of the device, and

wherein generating the control sequence occurs iteratively to refine the control sequence according to a quadratic optimization in stochastic sequential action control.

19. The method of claim 14 , wherein identifying the objects includes predicting trajectories of the objects according to a generative behavior prediction model out to a prediction time horizon, and

wherein generating the control sequence is a sampling-based stochastic model predictive control approach that is risk aware.

20. The method of claim 14 , wherein identifying the objects includes determining an initial state of the device, including dynamics and a current location, a reference trajectory indicating a desired path of the device, and a nominal control schedule for controlling the device over the reference trajectory.