A Data-Driven Krasovskii-Based Approach for Safety Controller Design of Time-Delayed Uncertain Polynomial Systems
summary
The gist
We develop a data-driven framework for the synthesis of robust Krasovskii control barrier certificates (RK-CBC) and corresponding robust safety controllers (RSC) for discrete-time input-affine…
In short
This work develops a data-driven method to create safety controllers for uncertain, time-delayed polynomial systems without knowing the exact system model. By using only observed input-state data from a finite period, it synthesizes robust Krasovskii control barrier certificates and safety controllers. This allows for formal safety guarantees over an infinite time horizon despite unknown disturbances and delays.
Key concepts
- RK-CBC
- The Robust Krasovskii Control Barrier Certificate is a mathematical condition used to prove system safety. It is specifically designed to handle systems with time delays by aggregating the delay effects into the certificate's structure, ensuring that trajectories stay outside an unsafe region even when dynamics are uncertain.
- Data-Driven Synthesis
- This approach builds the safety controller directly from observed input-state data collected over a finite time period. Instead of needing a full mathematical model beforehand, the method uses this real data to synthesize the necessary safety conditions, bypassing traditional identification steps.
- Sum-of-Squares (SOS) Optimization
- This is a mathematical tool used to solve the synthesis problem. It transforms complex safety constraints into a tractable optimization program. The authors use SOS to find the parameters for the control barrier certificate that guarantee robustness against disturbances and delays.
- Input-Affine Polynomial System with Delays (dtIAUPS-td)
- This describes a type of system where the next state depends on current inputs, current states, and delayed states. The key challenge is that the system matrices defining this relationship are unknown, yet safety must be guaranteed over time despite these unknown dynamics.
Terminology used across episodes
This episode discusses
- A Data-Driven Krasovskii-Based Approach for Safety Controller Design of Time-Delayed Uncertain Polynomial Systems · Paper Radio
- Compositional Design of Safety Controllers for Large-Scale Stochastic Hybrid Systems
- From Formal Methods to Data-Driven Safety Certificates of Unknown Large-Scale Networks
- Data-Driven Safety Certificates of Infinite Networks with Unknown Models and Interconnection Topologies
- LMI Properties and Applications in Systems, Stability, and Control Theory
The paper
A Data-Driven Krasovskii-Based Approach for Safety Controller Design of Time-Delayed Uncertain Polynomial Systems · Read on arXiv
Omid Akbarzadeh, MohammadHossein Ashoori, Amy Nejati, Abolfazl Lavaei
Newcastle University
Transcript
Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.
Rosa: We're looking at the paper: "A Data-Driven Krasovskii-Based Approach for Safety Controller Design of Time-Delayed Uncertain Polynomial Systems".
Dev: Here's the gist of the paper. We develop a data-driven framework for the synthesis of robust Krasovskii control barrier certificates (RK-CBC) and corresponding robust safety controllers (RSC) for discrete-time input-affine uncertain polynomial systems with unknown dynamics, while explicitly accounting for unknown-but-bounded disturbances and time-invariant delays using only observed input–state data Problem Statement The central problem addressed is the design of an RK-CBC and its associated R-SC for a discrete-time input-affine polynomial system with unknown dynamics, unknown but bounded disturbances, and time-invariant delays using only observed input–state data.
Rosa: This framework aims to establish robust safety guarantees over an infinite time horizon without requiring an explicit mathematical model of the system.
Dev: The difficulty stems from the system's mathematical model being unavailable and the need to explicitly incorporate time delays into safety conditions while remaining robust to unknown disturbances.
Rosa: Data-Driven Methodology The proposed approach relies solely on input–state data collected over a finite time horizon to directly synthesize the RK-CBC and R-SC without needing an explicit system model.
Title and authors: Rosa: So we're looking at a paper called "A Data-Driven Krasovskii-Based Approach for Safety Controller Design of Time-Delayed Uncertain Polynomial Systems" by Omid Akbarzadeh, MohammadHossein Ashoori, Amy Nejati, and Abolfazl Lavaei. It sounds really technical.
Dev: Yeah, it tackles the problem of designing a robust safety controller when you don't know the system dynamics at all. The title tells you immediately it’s data-driven and involves Krasovskii certificates for systems with delays and uncertainties.
Rosa: Exactly, what's important here is that they are doing this using only observed input state data. They aren't trying to model the whole system first, which is a huge deal when you don't have a mathematical model available.
Taro: From my side, I’m curious how robust it actually is when the world misbehaves, especially since we're dealing with unknown disturbances and time-invariant delays.
Dev: That’s the main challenge they set out to solve in their abstract. The core difficulty is that you need a safety condition that explicitly accounts for those time delays during the design process while staying safe against unknown disturbances, all without knowing the actual system dynamics.
Rosa: It seems like they are focusing on making this work using only input state data collected over a finite time horizon, which is something I can relate to my own field robotics work.
The paper's summary: Rosa: So what’s the actual method? Basically, they use that input state data to directly synthesize two things: a robust Krasovskii control barrier certificate and a corresponding robust safety controller. It’s not about building a model first, it’s about synthesizing the safety structure itself from the data.
Dev: They formulate this synthesis as a sum-of-squares optimization program, which is how they make it computationally manageable. The key improvement they introduce is putting those delay terms directly into the certificate construction by aggregating them in a specific way.
Taro: Aggregating them sounds complicated when you’re trying to guarantee safety over an infinite time horizon, especially with unknown dynamics involved. How do you keep that guarantee tight?
Rosa: The paper claims that if these conditions are met, any trajectory starting from a safe set will stay outside the unsafe region forever. This is the safety guarantee they establish for this data-driven framework.
Dev: That infinite horizon safety is what’s significant because it means you don't need to worry about the system changing its behavior in ways you haven't seen in your data collection period, as long as those disturbances and delays stay bounded.
Rosa: So, they are using a quadratic matrix form involving some matrices He and Se to define these conditions for the certificate. That’s where the specific mathematical structure comes in.
The paper's improvements: Dev: The main technical improvement they point to is moving from prior work on unknown systems with delays, which often needed accurate system knowledge, to this data-driven approach that works purely from observed data.
Taro: What else are they improving? I'm thinking about the complexity of the math itself. If you’re synthesizing a controller, you want it to be fast enough for real-time loops.
Rosa: They claim that by structuring the problem this way, they reduce the set-product complexity from something like an h plus one fold product down to at most a two-fold product in the final certificate conditions. That makes solving the optimization program much faster.
Dev: That reduction in complexity is really important for practical implementation because it lowers the computational burden during that synthesis step, which is often a bottleneck when you're working with these kinds of certificates.
Taro: So, even though they are dealing with time delays and disturbances, this data-driven method makes the final safety condition much easier to calculate than before.
Conclusion: Rosa: So to wrap up on this paper titled "A Data-Driven Krasovskii-Based Approach for Safety Controller Design of Time-Delayed Uncertain Polynomial Systems," the main implication is that you can synthesize a robust safety controller even when you have zero knowledge of the system dynamics, provided you can collect some input state data.
Dev: That’s right, the paper shows how to handle unknown dynamics, time delays affecting both states and inputs, and bounded disturbances all at once in a way that guarantees infinite time horizon safety using only observed data.
Taro: For someone on the ground testing these systems, this means you don't need a perfect mathematical map of what’s happening inside the machine to build a safe control law; you just need good input state measurements.
Rosa: And for me, it means we can move away from slow identification phases and jump straight to safety certification using empirical data instead of waiting months for model fitting.
Dev: It gives us a way to formally guarantee safety under those specific uncertainty conditions without needing an explicit mathematical model of the system, which is a big step forward for real-time control design.
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