US 11,880,188 B2 · Intelligent distribution of data for robotic and autonomous systems

Abstract

The present disclosure relates to the intelligent distribution of data for robotic, autonomous, and similar systems. To reduce the impact of multi-agent coordination on networked systems embodiments are disclosed that include the use of action-based constraints which yield constrained-action POMDP (CA-POMDP) models, and probabilistic constraint satisfaction for the resulting infinite-horizon finite state controllers. To enable constraint analysis over an infinite horizon, an unconstrained policy is first represented as a finite state controller (FSC). A combination of a Markov chain Monte Carlo (MCMC) routine and a discrete optimization routine can be performed on the finite state controller to improve probabilistic constraint satisfaction of the finite state controller, while minimizing impact to a value function.

Claims20

Dependent claims

A computer-implemented method for implementing a constrained-action partially-observable Markov decision process (CA-POMDP), comprising:

generating, by at least one computing device, an unconstrained policy through execution of a partially observable Markov decision process (POMDP) routine;
generating, by the at least one computing device, an optimal constrained policy by invoking a policy iteration routine that applies at least one transformation to a finite state controller, the optimal constrained policy generated comprising an optimal finite state controller having a threshold probability of operating within at least one operational constraint within a period of time;
applying, by the at least one computing device, a probabilistic constraint satisfaction routine to bring the optimal finite state controller within a probabilistic constraint satisfaction threshold;
and
determining, by the at least one computing device, to perform an action based at least in part on a current state of the at least one computing device and the optimal finite state controller.

2 more elements

A system for implementing a constrained-action partially-observable Markov decision process (CA-POMDP), comprising:

at least one computing device;
and
program instructions executable in the at least one computing device that, when executed, direct the at least one computing device to:
generate an unconstrained policy through execution of a partially observable Markov decision process (POMDP) routine;
generate an optimal constrained policy by a discrete and combinatorial optimization routine that applies at least one transformation to a finite state controller, the optimal constrained policy generated comprising an optimal finite state controller having a threshold probability of operating within at least one operational constraint within a period of time;
apply a probabilistic constraint satisfaction routine to bring the optimal finite state controller within a probabilistic constraint satisfaction threshold;
and
determine to perform an action based at least in part on a current state of the at least one computing device and the optimal finite state controller.

5 more elements

Dependencies are inferred from the claim text.

Intelligent distribution of data for robotic and autonomous systems

Life of this patent

Intelligent distribution of data for robotic and autonomous systems

Life of this patent

Abstract

Claims20