Communication-Aware Synthesis of Safety Controller for Networked Control Systems

Networked control systems (NCS) are widely used in safety-critical applications, but they are often analyzed under the assumption of ideal communication channels. This work focuses on synthesizing safety controllers for discrete-time linear systems affected by unknown disturbances operating in imperfect communication channels, guaranteeing safety by constructing ellipsoidal robust safety invariant (RSI) sets and verifying their invariance through linear matrix inequalities (LMI).

The gist

The proposed method guarantees safety by constructing ellipsoidal robust safety invariant (RSI) sets and verifying their invariance through linear matrix inequalities (LMI), which are formulated and solved as semi-definite programming (SDP), simultaneously considering controller synthesis and communication errors without requiring explicit modeling of the communication channel.

System Model and Error Characterization

The paper considers a discrete-time linear system described by the equation:

x(k + 1) = Ax(k) + Bu(k) + d(k), k ∈ N, where A is the system matrix, B is the input matrix, x(k) is the state of the system, u(k) is the control input, and d(k) denotes external disturbance. The control input is defined as u(k):= Kxˆ(k), where xˆ(k) is the estimated state produced by a Kalman filter influenced by communication channels and estimation errors. Due to imperfections like packet drops or time delays, an uplink communication error eup(k) is introduced, characterized by eup(k):= ˆx(k) − x(k). Substituting these into the system dynamics generates the closed-loop equation: x(k + 1) = (A + BK)x(k) + BKeup(k) + d(k). This leads to defining the system state error induced by imperfect communication and state estimation as e(k):= BK eup(k), which belongs to the bounded set ∆(ε), where ε is the system state error bound.

Communication Error Bound Analysis

The objective of this section is to derive a system state error bound ε based on the uplink communication error bound εup. The analysis begins by defining a Kalman filter structure with process noise Q and measurement noise R, and defines the conservative process uncertainty covariance Q¯:= Q + γIn. Theorem 1 establishes that the error covariance Pk of the Kalman filter has an upper bound Pk ⪯ p¯kIn, where p¯k is recursively defined. By applying a standard notion of exponential boundedness (Definition 2), Theorem 2 provides a computable bound for the uplink communication error: εup = (¯p0g¯ squared + q¯ 1 − g¯ 2)χ 2n, 1−δ, where δ is a design parameter. This bound is independent of real-time measurement data and depends on the closed-loop matrix G = A + BK.

System State Error Bound Derivation

The system state error bound ε is then computed by incorporating the uplink communication error bound εup into the closed-loop error dynamics (8). Theorem 3 states that if ¯g squared ≤ 1, then e(k)⊤e(k) ≤ ε for all k ∈ N, where ε = (¯g + A2) 2εup. This bound is computable by solving a semi-definite programming (SDP) problem. The key insight is that the coupling between the controller and the state error is handled by setting ¯g:= √κρ, with κ ∈ (0, 1] and ρ ∈ [1, 1/κ], ensuring that the condition for boundedness in Theorem 2 is satisfied.

LMI-Based Controller Synthesis

The core of the framework involves translating invariance conditions into a convex optimization problem (OP) using LMI constraints. The goal is to design a safety envelope S = M−1 and a state-feedback controller u = F L−1xˆ, where F:= KL. The optimization problem OP seeks to minimize-log(det(L)) subject to several constraints:

1. Contraction and Invariance: κL L⊤A⊤ + F⊤B A L + BF L ⪰ 0, and αI ≤ L (where α is defined based on γ, ε, and κ).

2. Set Inclusion: cjLc⊤j ≤ 1 for all state constraints cj.

3. Input Constraints: Constraints involving umax and the communication error set ∆(ε) are enforced through auxiliary variables τi and U, which are relaxed to ensure convexity by imposing U ⪯ λmin(L) 2W.

Co-design Framework and Case Study

The framework is implemented via Algorithm 1, which iteratively searches over contraction parameters κ and scaling parameters ρ to solve OP. This joint design achieves a "balanced tradeoff between communication efficiency and safety assurance.

Here are the specific improvements to AI systems derived from this scientific paper, focusing on integrating robust safety guarantees under imperfect communication:

1. Effective Safety Controller Synthesis for Networked Control Systems (NCS):

The primary improvement is the ability to design a state-feedback controller that formally guarantees system safety even when communication channels are unreliable (packet drops, time delays). This moves AI control systems from merely being stable to being provably safe.

2. Robustness Against Communication Imperfections:

AI systems will be capable of operating reliably in real-world cyber-physical environments where sensor data transmission is noisy or intermittent. The system can handle:

• Packet Loss and Time Delays: The controller explicitly accounts for delayed or missing state information by bounding the resulting estimation error.

• Bandwidth Limitations and Quantization Errors: By incorporating these as bounded uncertainties, the controller maintains safety margins even when communication bandwidth is constrained or data is quantized (as demonstrated in the cruise control case study).

3. Co-Design of Communication Efficiency and Safety:

The proposed framework allows for a simultaneous optimization of two conflicting goals: maintaining high safety assurance and minimizing communication overhead.

• The system can determine the optimal controller gain matrix (K) and the required safety region (M) that satisfies both state constraints and communication limits simultaneously, leading to a more efficient control law than methods that optimize them separately.

4. Formal Safety Guarantees via Robust Invariant Sets (RSI):

The core mechanism is the construction of an ellipsoidal Robust Safety Invariant Set (RSI set), defined by the LMI formulation in Problem 1.

• The AI controller's objective is to ensure that regardless of bounded external disturbances and the computed communication/estimation errors, the system state trajectory remains strictly within this pre-defined safe region. This provides a mathematically rigorous safety envelope rather than heuristic performance guarantees.

5. Adaptive Safety Margin Management:

The iterative search framework (Algorithm 1) allows the system to dynamically tune its robustness parameters based on communication uncertainty bounds.

• The AI controller can adapt the size and shape of its safety set (via M) and control gain (K) in response to the estimated level of communication imperfection, ensuring that the safety margin is precisely what is needed for a given network condition, leading to optimal use of resources.

6. System State Error Bounding:

The framework provides a computable upper bound on the actual system state error induced by the combination of communication errors and estimation inaccuracies.

• This allows for proactive monitoring: If real-time monitoring indicates that the current state error exceeds this derived bound, it signals an imminent safety violation before it occurs, enabling faster emergency maneuvers or communication recovery protocols.

7. Applicability to Complex Systems (Cruise Control Analogy):

The methodology is demonstrated on a complex system (truck/trailer dynamics) with multiple simultaneous channel imperfections (packet loss, quantization, delay).

• The improved AI controller can manage highly coupled systems where various communication failures occur concurrently without losing formal safety guarantees.