A local recursive least squares approach for discrete-time adaptive fuzzy control

summary

Video file (mp4)

The gist

A local recursive least squares approach for discrete-time adaptive fuzzy control proposes a membership-weighted RLS law with a forgetting factor to approximate unknown nonlinearities in quasi-Linear

In short

This work proposes a membership-weighted local recursive least squares with a forgetting factor to estimate unknown nonlinearities in discrete-time adaptive fuzzy control systems (qLPV/TS). The method uses rule-specific covariance to reduce memory and simplifies adaptation. It guarantees the closed-loop system remains uniformly bounded, providing LMI synthesis conditions for matched, sector-bounded, and norm-bounded nonlinearities.

Key concepts

Membership-weighted RLS with Forgetting Factor
This is an estimation strategy used to find the unknown parameters of a fuzzy model. It weights the least squares updates by how much each rule 'belongs' to a certain state (membership function). The forgetting factor helps the estimator track changing system dynamics over time, improving performance in adaptive control.
qLPV/Takagi–Sugeno (qLPV/TS) Systems
These are types of nonlinear systems that are modeled using fuzzy logic. In qLPV/TS models, the system's output is a weighted average of several linear functions. The paper uses these models to approximate complex, unknown nonlinearities in the physical system being controlled.
LMI Synthesis Conditions
Linear Matrix Inequalities (LMIs) are mathematical constraints used to prove stability and boundedness for adaptive systems. The paper derives three distinct sets of LMIs tailored for different types of unknown nonlinearities (matched, sector-bounded, or norm-bounded). These conditions ensure the adaptive controller will not become unstable.
Ultimate Uniform Boundedness
This means that despite the unknown nonlinearities in the system, the error between the actual system behavior and what the adaptive controller predicts will stay within a finite, predictable range. The analysis shows this boundedness is guaranteed when specific adaptation conditions are met.

Terminology used across episodes

This episode discusses

The paper

A local recursive least squares approach for discrete-time adaptive fuzzy control · Read on arXiv

V´ıctor Costa da Silva Camposa, Mariella Maia Quadrosb

Electronics Engineering Department, Universidade Federal de Minas Gerais · Instituto Federal de Educação, Ciência e Tecnologia de Minas Gerais

This paper proposes a local recursive least squares (RLS) estimation strategy for discrete-time adaptive fuzzy control of nonlinear systems represented in quasi-Linear Parameter Varying (qLPV)/Takagi--Sugeno (TS) form. Unknown nonlinear terms are approximated by a constant-consequent TS fuzzy model, and the consequent parameters are updated by a membership-function-weighted RLS law with a forgetting factor. The proposed estimator keeps a different covariance for each rule, considerably reducing the memory footprint of the least-squares updates. The adaptation is also simplified since each rule is only adapted when it is active. From these properties, we are capable of showing that the adaptation law ensures bounded local adaptation errors. Building upon this adaptation law, Linear Matrix Inequality (LMI) synthesis conditions are presented for matched, sector-bounded and norm-bounded unknown nonlinearities, guaranteeing ultimate uniform boundedness of the adaptive control system in closed loop. Three numerical examples are presented to illustrate the adaptive control conditions in the three cases: a planar manipulator with unknown gravity direction, a two-tank system with unknown coupling, and a Brushless DC (BLDC) motor acting as thrust for an efficiency vehicle.

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: I'm Rosa, and with me are Dev and Taro, guest researcher.

Dev: Today's paper: "A local recursive least squares approach for discrete-time adaptive fuzzy control".

Rosa: A local recursive least squares approach for discrete-time adaptive fuzzy control proposes a membership-weighted RLS law with a forgetting factor to approximate unknown nonlinearities in quasi-Linear Parameter Varying/Takagi–Sugeno (qLPV/TS) systems,…

Dev: First, who's behind it and why it matters.

Title and authors: Rosa: So, we're looking at a paper titled "A local recursive least squares approach for discrete-time adaptive fuzzy control," and the authors are V´ıctor Costa da Silva Campos and Mariella Maia Quadros. It sounds like they’re tackling how to make control systems smart when the dynamics aren't perfectly known.

Dev: Yeah, I see that title points right toward using a local recursive least squares method for discrete-time adaptive fuzzy control, which suggests a focus on real-time adaptation within a specific system structure.

Taro: From an autonomy standpoint, it seems like they are trying to build controllers that can handle situations where the environment or internal dynamics change unexpectedly during operation.

Rosa: Exactly, and the implications here are pretty big because if this works outside of a controlled lab setting, it means we could deploy systems in much more unpredictable real-world scenarios than we currently manage.

Dev: It’s about moving beyond just robust controllers that have fixed limits; this paper is proposing a mechanism where the controller can actively estimate and adjust its parameters based on what it sees.

The paper's summary: Rosa: So, what the core idea of this work is, based on the summary provided, they’re using a membership-weighted local recursive least squares with a forgetting factor to approximate unknown nonlinearities in systems modeled in quasi-Linear Parameter Varying or Takagi–Sugeno fuzzy form.

Dev: That means they are taking these fuzzy models where the consequent parameters are unknown and using this RLS approach to find those parameters while accounting for the membership functions of each rule.

Taro: And what's interesting is that they simplify things by only adapting each rule when it’s active, which cuts down on computational load significantly.

Rosa: Right, and they also keep a different covariance for each rule, which the summary says considerably reduces the memory footprint of the least-squares updates because it only adapts when necessary.

Dev: That sounds like a practical improvement for deployment; managing memory is crucial when you're running complex loops in an embedded system.

The paper's improvements: Rosa: Now, looking at what they actually improved, the paper suggests a membership-weighted local recursive least squares with a forgetting factor approach that estimates consequent parameters for a constant-consequent TS fuzzy model.

Dev: They detail specific adaptation laws, like equation (eight) and (nine), which show how the parameter estimates pi(k+one) and theta i(k+one) are updated based on the previous values and some gain terms.

Taro: That part about each rule only being adapted when it is active is a key improvement because it simplifies the adaptation process, making the control loop more efficient.

Rosa: And they go deeper into the covariance dynamics too; they show that whenever hik isn't zero, wik converges monotonically to zero, which in turn implies that whenever hik isn't zero, p ik converges monotonically to one over alpha.

Dev: That monotonic convergence of the covariances and parameters is a strong indicator of stability for the estimation part of the system.

Conclusion: Rosa: So, to wrap up what we’ve discussed about this paper, it boils down to proposing this local recursive least squares approach for discrete-time adaptive fuzzy control, which aims to approximate unknown nonlinearities in qLPV/TS systems.

Dev: The main implication is that they've derived LMI synthesis conditions that guarantee the ultimate uniform boundedness of the closed loop adaptive system concerning the approximation error for matched, sector-bounded, and norm-bounded unknown nonlinearities.

Taro: For autonomy, this means we could have controllers that are adaptive enough to handle unpredictable world behavior without needing a perfectly modeled environment upfront.

Rosa: It’s quite robust because they cover three different types of nonlinearity—matched, sector-bounded, and norm-bounded—providing different LMI conditions for each case.

Dev: The paper does present a separate feedforward condition specifically for the norm-bounded case to approximately render a desired output immune to it, which is interesting.

Taro: That capability to handle uncertainty across these different nonlinearity types really expands what we can expect from adaptive control systems in dynamic environments.

More episodes

← Home