Strong nonlinear detectability and moving horizon estimation for nonlinear systems with unknown inputs
summary
The gist
Strong nonlinear detectability and moving horizon estimation for nonlinear systems with unknown inputs considers state estimation for general nonlinear discrete-time systems subject to measurement…
In short
The paper introduces strong nonlinear detectability as a concept necessary and sufficient for designing unknown input state estimators (UISEs) for nonlinear systems with measurement noise and potentially unbounded inputs. It proposes a two-stage Moving Horizon Estimation (MHE) scheme using reduced-order models to achieve better estimation accuracy while significantly reducing online computational complexity compared to full-order MHE methods.
Key concepts
- Strong Nonlinear Detectability
- This is a condition that determines whether an unknown input state estimator can exist for a nonlinear system. It is defined broadly, considering past and current measurements, and it serves as the fundamental requirement for the existence of such estimators.
- Moving Horizon Estimation (MHE)
- MHE is an estimation technique used to reconstruct the current state of a system by optimizing over a future time horizon. The paper uses MHE to design UISEs, which allow systems to estimate states even when unknown inputs are unbounded and measurement noise is present.
- Reduced-Order Model Transformation
- To improve efficiency, the authors use coordinate transformations (diffeomorphisms) to create a simpler, reduced-order model. This transformation is linked to a tighter condition for strong nonlinear detectability, enabling the design of a two-stage estimator with fewer optimization variables.
Terminology used across episodes
This episode discusses
- Strong nonlinear detectability and moving horizon estimation for nonlinear systems with unknown inputs · Paper Radio
The paper
Strong nonlinear detectability and moving horizon estimation for nonlinear systems with unknown inputs · Read on arXiv
Chemnitz University of Technology · Universidad Nacional Autonoma de México
Transcript
Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.
Rosa: We're looking at the paper: "Strong nonlinear detectability and moving horizon estimation for nonlinear systems with unknown inputs".
Dev: Here's the gist of the paper. Strong nonlinear detectability and moving horizon estimation for nonlinear systems with unknown inputs considers state estimation for general nonlinear discrete-time systems subject to measurement noise and possibly unbounded unknown inputs The main contributions are threefold We propose a concept of strong nonlinear detectability and provide its equivalent Lyapunov-like characterization We present an analysis and design framework for UISEs built on a moving horizon estimation scheme using full-order models By leveraging a tighter condition for strong nonlinear detectability, we derive a reduced-order model that enables the design of a two-stage MHE-based UISE, reducing the online computational complexity compared to the MHE-based UISE using full-order models Concept of Strong Nonlinear Detectability The paper introduces strong nonlinear detectability as a concept sufficient and necessary for the existence of unknown input state estimators (UISEs) This notion is defined in Definition 3.1, which incorporates not only past noise and output measurements but also current ones, which makes it more general than notions formulated solely with past noise and measurements The equivalence between strong nonlinear detectability and the existence of a UISE is established through Lyapunov-like characterizations Specifically, Theorem 3.4 states that there exists a UISE if and only if the system is strongly nonlinearly detectable.
Rosa: MHE-based UISE Design Framework The design framework involves two main approaches for UISEs 1.
Dev: MHE-based UISE using full-order models: This scheme guarantees convergent estimation errors for convergent measurement noises even when unknown inputs are unbounded, extending robust MHE developments to nonlinear systems with unbounded disturbances However, this approach uses only past measurements to estimate the current state and follows a one-step ahead estimation strategy.
Title and authors: Rosa: So the core idea here is that strong nonlinear detectability provides a condition, like a rule, that lets us know if an unknown input state estimator can even exist for our system.
Dev: They define this strong nonlinear detectability and then show it's equivalent to the existence of a UISE, which means we can reconstruct the true state even with those big unknown inputs.
Rosa: And they then use this new concept to build an entire framework for designing UISEs by using a moving horizon estimation strategy with a full-order model that takes both past and current measurements.
Dev: That full-order model version is important because it promises convergent error estimates even when the unknown inputs are unbounded, which extends some existing robust moving horizon estimation work to these nonlinear systems.
Rosa: But there’s a catch, they point out that this traditional method only uses past measurements for the current state estimate, which makes it a one-step ahead strategy.
Dev: That’s where the paper gets interesting because they suggest we can tighten that strong nonlinear detectability condition to design something much more efficient.
Rosa: They use this tighter condition to create a reduced-order model, which lets them build a two-stage MHE based UISE that's computationally less expensive than the full-order version.
The paper's summary: Dev: The main improvement they propose is moving away from the full-order model to a two-stage approach that uses a reduced-order model, which really tackles the computational complexity issue.
Rosa: They achieve this by linking strong nonlinear detectability with the detectability of these smaller reduced-order models using a coordinate transformation, which involves something called a diffeomorphism T of x.
Dev: That transformation helps them decouple the system into parts that are easier to handle, and assumption four point two ensures there’s a function psi that makes this decoupling possible.
Rosa: So instead of running the full-order MHE scheme every time, you first get a one-step ahead estimate from the reduced model and then use the current measurement to compute the final current state estimate.
Dev: The result is that this two-stage UISE is actually more computationally efficient than conventional moving horizon estimators, which they show in simulations of a plant growth process.
Rosa: And in those simulations, they found that the estimation error for their proposed UISEs was better than what you'd get from the standard MHE or other methods when dealing with noisy data.
The paper's improvements: Dev: So, to wrap up, this paper on "Strong nonlinear detectability and moving horizon estimation for nonlinear systems with unknown inputs" establishes a condition for existence that lets us build UISEs even when the inputs are unbounded.
Rosa: And they show how leveraging that condition through a two-stage reduced-order model can lead to an estimation scheme that is both more accurate and less computationally demanding than the full-order approach.
Dev: The implication is we have a way to handle these complex nonlinear systems with unknown inputs by using this strong detectability concept and then designing estimators that don't require solving the full problem every time.
Rosa: For those of us on the outside, it means we can design better observers for real-world robotics or control problems where we just can't bound everything perfectly.
Taro: I wonder what happens when the world misbehaves really badly, like a huge unknown input suddenly appears; does this two-stage approach handle that gracefully?
Dev: That’s a good question, Taro. The paper focuses on the estimation error convergence under those conditions, and it suggests it can isolate the unknown inputs better than other schemes.
Rosa: And I'm curious how long these estimators stay stable in practice when you run them for a long time compared to the conventional MHE methods they compare against.
Dev: The paper shows that the two-stage UISE has smaller optimization variables, which directly translates to faster online computation time during real-time operation.
Taro: So it’s about getting better accuracy without bogging down our real-time loops, right?
Conclusion: Rosa: So we’re wrapping up on this paper about "Strong nonlinear detectability and moving horizon estimation for nonlinear systems with unknown inputs." Basically, they show you a new way to know if an estimator even exists for your system when you have those unpredictable unknown inputs.
Dev: Right. They introduce this concept of strong nonlinear detectability and prove that it’s the exact condition needed for an unknown input state estimator to work.
Rosa: It’s not just about knowing the past measurements, though; they make it more general by including what's happening right now too, which is a big deal for real systems.
Dev: Exactly. And they show that this strong detectability condition is equivalent to being able to build an unknown input state estimator.
Taro: I always wonder if this works outside the lab conditions we test in, or if it’s just theoretical perfection on paper.
Rosa: Well, the authors are looking at how these estimators perform when there's noise in the measurements and those big unknown inputs that aren't bounded.
Dev: They present a design framework using moving horizon estimation with a full-order model first, which handles those unbounded inputs but uses only past measurements for the current state estimate.
Rosa: Then they introduce this two-stage scheme where you use a reduced-order model to get a fast one-step ahead estimate and then correct it with the current measurement.
Dev: That reduced-order part is what’s interesting because it cuts down the computational load compared to running the full-order model all the time.
Taro: So if you can reduce complexity, does that mean we can run this on something smaller, like a drone or a wearable device?
Rosa: That’s one implication. Because they show in their simulations with a crop growth process that this two-stage scheme actually gives better accuracy than the standard moving horizon estimators.
Dev: It's not just theoretically sound; the comparison shows the error is smaller when you use this method, even with noise in the measurements.
Taro: I’m still thinking about what happens when those unknown inputs become really erratic, like a sudden huge disturbance. How gracefully does this system recover?
Rosa: That’s future research they mention—jointly estimating both the state and those unknown inputs together. They also plan to look at non-asymptotic estimators using moving horizon estimation.
Dev: It sounds like they are focusing on making these estimators practical for real-time control, even when the system dynamics are nonlinear and messy.
Rosa: So, for anyone else listening, this paper is about giving us a solid mathematical foundation to design better observers that can handle those tricky unknown inputs in real systems.
Dev: It’s a lot of work showing the equivalence of detectability concepts and then designing efficient estimators on top of it all.
Taro: Anyway, next time we talk about estimation, we'll look at how data-driven methods like the Koopman operator can handle these kinds of nonlinear systems.
More episodes
- 2608.13438-ContactGuard: Pre-Contact Execution Monitoring with Action-Conditioned Latent World Models
- 2610.34593-Future Prediction as Training-Only Video Supervision for Pretrained VLA Policies
- 2610.35385-EmbodiedLab: Beyond Manipulation, Benchmarking Embodied Agents That Act to Know
- 2610.36378-Data-Driven Hamiltonian for Neural Optimal Control
- 2610.37275-A Heuristic for Scheduling Autonomous Vehicles at Intersections
- 2610.37288-Dynamic Coordination of Autonomous Vehicles at Intersections
- 2610.37632-Real-Time Coordination of Autonomous Vehicles at Intersections
- 2611.13779-Softness is something you also need! Why compliance helps control: information metrics of a passive walker with compliant ankle
- 2611.13411-Residual-Corrected Physics-Informed Neural Network KKL Observers with Input-to-State Stability Guarantees
- 2611.14488-Is Your Robot's Memory Just a Clock? Temporal Shortcuts in Memory-Augmented VLA Policies