Anytime Control using Input Sequences with Markovian Processor Availability

TL;DR

This paper proposes an anytime control algorithm that buffers sequences of control inputs based on Markovian processor availability, ensuring stochastic stability despite time-varying computational resources.

Contribution

It introduces a novel control method that accounts for Markovian processor availability, providing stability guarantees under uncertain computational conditions.

Findings

Algorithm maintains stability with Markovian resource fluctuations.

Buffering control sequences improves robustness against resource variability.

Provides Lyapunov-based conditions for stochastic stability.

Abstract

We study an anytime control algorithm for situations where the processing resources available for control are time-varying in an a priori unknown fashion. Thus, at times, processing resources are insufficient to calculate control inputs. To address this issue, the algorithm calculates sequences of tentative future control inputs whenever possible, which are then buffered for possible future use. We assume that the processor availability is correlated so that the number of control inputs calculated at any time step is described by a Markov chain. Using a Lyapunov function based approach we derive sufficient conditions for stochastic stability of the closed loop.