Iterative Learning Control with Passive Incomplete Information: Algorithms Design and Convergence Analysis

In iterative learning control (ILC), a system improves its performance by using information from earlier repetitions of a task. But what happens when those records are incomplete—because measurements disappear, messages arrive late or out of order, or an operation ends before the full trial is recorded? This monograph develops algorithms and convergence analysis for precisely those practical conditions.

Dong Shen organizes the discussion around increasingly complex forms of incomplete information, connecting mathematical control theory with the communication limits encountered in networked systems.

Three models for missing data

The first part studies one-sided data dropouts through random-sequence, Bernoulli-variable, and Markov-chain models. These frameworks are applied to linear and nonlinear systems, including systems with unknown control direction. The book analyzes intermittent and successive update schemes and establishes convergence results for the settings considered.

When both sides of the control loop are affected

Dropouts at both the measurement and actuator sides create an additional challenge: the control signal calculated by the learning algorithm may differ from the signal that actually reaches the system. The monograph treats this asynchronism in deterministic, stochastic, and nonlinear systems, developing analyses suited to each case.

Beyond data dropouts

The final part broadens the discussion to communication delays, transmission disorder, limited memory, randomly varying iteration lengths, and large-scale systems with asynchronous subsystems. Across these topics, algorithm design is paired with proofs and illustrative simulations, giving readers a way to follow both the proposed methods and the reasoning behind their convergence.

For readers working with control theory

Written for students, engineers, and researchers in iterative learning control, this focused technical treatment is especially relevant to work in stochastic control and networked systems. Its progression—from basic ILC formulations to multiple communication constraints—helps situate individual algorithms within a wider framework for learning under incomplete information.