Event-Triggered Funnel Control for Output Reference Tracking of Unknown Nonlinear Systems

Step 1: We show the existence of a maximal solution (y,I):[0,\omega) \rightarrow \mathds{R}^n \times \mathds{R}^n.
We will inductively show the existence of a solution on the intervals on which S_Mu is constant. Define the non-empty open set

Define the function

for some fixed v \in \mathds{R}^n. Then, the initial value problem (?), (?) takes the form

on an interval [s,\tau) such that there are no points of discontinuity of S_Mu in this interval, with v = S_Mu(s). We observe that G_u is locally Lipschitz in (y,I), and locally integrable in t. Whenever (s, y^{s}, I^{s})\in \mathcal{D}, application of [Walt98, § 10, Thm. XX] yields the existence of a maximal solution (y,I):[s,\tilde\omega) \rightarrow \mathds{R}^n \times \mathds{R}^n for some \tilde\omega \in (s,\infty]. Moreover, the closure of the graph of the maximal solution is not contained in a compact subset of \mathcal{D}.

If \left\| u(t) - v \right\|_{ } < M for all t \in [s,\tilde\omega) where u is defined as in (?), then \frac{\left\| e(t) \right\|_{ }}{1-\varphi(t)^2\left\| e(t) \right\|_{ }^2} \leq \left\| v \right\|_{ } + M for all t \in [s,\tilde\omega). This implies that u(\cdot) is uniformly bounded on this interval. Hence, either \tilde{\omega} = \infty, or the solution is contained in a compact subset of \mathcal{D}, which is a contradiction. This shows that either \tilde\omega = \infty, or there exists t \in [s,\tilde\omega) such that \left\| v - u(t) \right\|_{ } \geq M.

We inductively define the sequence (t_k)_{k \in T} for some index set T \subseteq \mathds{N} as follows. Set t_0 = 0, and observe that (t_0,y^0,I^0) \in \mathcal{D}. For each k \in \mathds{N}_0, apply the above argument for the existence of a maximal solution on some interval [t_k,\tilde\omega). If \left\| u(t) - u(t_k) \right\|_{ } < M for all t \in [t_k,\tilde\omega), then \tilde\omega = \infty by the previous argument, and the induction terminates with T = \{ 1, \ldots, k\}. Otherwise, choose t_{k+1} \coloneq \inf \{ t\in [t_k,\tilde\omega) \mid \left\| u(t) - u(t_k) \right\|_{ } = M \}, which is well-defined by the previous argument and the continuity of u. Moreover, we observe that (t_{k+1},y(t_{k+1}),I(t_{k+1})) \in \mathcal{D} since t_{k+1} < \tilde\omega, closing the induction step. Set T = \mathds{N} if the induction did not terminate at any step k \in\mathds{N}.

By construction, the sequence (t_k)_{k \in T} is the sequence of discontinuities of S_Mu, and the closed-loop system (?), (?) coincides with (Proof) on the intervals [t_k,t_{k+1}) with v=u(t_k).

Hence, we can inductively conclude that there exists a maximal solution of the closed-loop system (?), (?) on [0,\omega) for \omega = \lim_{k \rightarrow \infty} t_k, if T = \mathds{N}, and \omega = \infty if \vert T \vert < \infty. Note that, at this point, \omega< \infty is not excluded in the case T=\mathds{N}.

Step 2: We show boundedness of the displacement I.
Aiming for a contradiction, assume that there is \tau_0 \in (0,\omega) such that \left\| I(\tau_0) \right\|_{ } > \max \{ \frac{M}{\alpha}, \left\| I^0 \right\|_{ } \} =: \tilde{M}. Choose \tau^{\star} \coloneq \sup \{ t \leq \tau_0 \mid \left\| I(t) \right\|_{ } = \tilde{M} \}, which is well-defined by continuity of I and the choice of \tilde{M}. Since by construction of the operator S_M, the estimate

holds, we can conclude

for all t \in [\tau^{\star},\tau_0]. Integration yields

a contradiction. Hence, we can conclude that \left\| I(t) \right\|_{ } \leq \tilde{M} holds for all t \in [0,\omega).

Step 3: We show \varphi(t) \left\| e(t) \right\|_{ } < \varepsilon and boundedness of u. Due to its continuity, f is uniformly bounded on each compact subset of \mathds{R}^n. We define F:\mathds{R}_{\geq 0} \rightarrow \mathds{R}_{\geq 0}, r \mapsto \left \Vert f_{\vert B_{r+\left\Vert \frac{1}{\varphi} \right\Vert_{\infty} + \Vert y_{\text{ref}} \Vert_{\infty}}(0)} \right\Vert_{\infty}, where B_r(0) \subseteq \mathds{R}^n denotes the ball of radius r centered at the origin. Then by construction, the estimate

holds, where \tilde{M} is defined in the previous step. Choose the constant \varepsilon \in (0,1) depending on the system parameters such that \varphi(0)\left\| y^0-y_{\text{ref}}(0) + I^0 \right\|_{ } < \varepsilon and

Assume that there exists s_1 \in [0,\omega) such that \varphi(s_1) \left\| e(s_1) \right\|_{ } > \varepsilon. Then s_0 \coloneq \sup \{ s \in [0,s_1] \mid \varphi(s)\left\| e(s) \right\|_{ } = \varepsilon \} is well-defined by the continuity of e and \varphi, as well as the choice of \varepsilon. Then on [s_0,s_1], the estimate

holds. Integration yields

a contradiction. We can conclude that \varphi(s)\left\| e(s) \right\|_{ } \leq \varepsilon for all s \in [0,\omega). Moreover, this yields \left\| u(t) \right\|_{ } \leq \frac{1}{1-\varepsilon^2} \left\Vert \frac{1}{\varphi} \right\Vert_{\infty}, hence u remains bounded.

Step 4: We derive a uniform lower bound on the length of the sampling intervals.
We can observe that, using \left\| I(t) \right\|_{ } \leq \tilde{M} on [0,\omega) from Step 2 and \left\| u(t) \right\|_{ } \leq \frac{1}{1-\varepsilon^2} \left\Vert \frac{1}{\varphi} \right\Vert_{\infty} from Step 3, we can obtain the estimate

This yields

for each k \in \mathds{N}, hence

In particular, this estimate ensures t_k \rightarrow \infty as k \rightarrow \infty.

Step 5: We show the existence of a global solution.
Assume that \omega < \infty. Then T = \mathds{N}, and by Step 1, \omega = \lim_{k \rightarrow \infty} t_k = \infty, due to the uniform lower bound on the interval lengths t_{k+1} - t_k established in Step 4. We can conclude \omega = \infty, finishing the proof.

We derive a lower bound on \frac{\tilde{M}}{F(\tilde{M})} by fixing an arbitrary M>0 and choosing a sufficiently small \alpha > 0 satisfying \left\| I^0 \right\|_{ } \alpha \leq M, implying \tilde{M} = \frac{M}{\alpha}. For a fixed M > 0, the dependence of \varepsilon on the choice of \alpha can be quantified via

by choosing equality in (?) for a suitable choice of initial values, where b = 1 + \left\Vert \frac{\dot{\varphi}}{\varphi} \right\Vert_{\infty} + \Vert \varphi \Vert_{\infty} (\Vert \dot{y}_{\text{ref}} \Vert_{\infty} + M). Hence, the dependence of the coefficient c(\tilde{M}) defined in (?) can be determined via

for some c_1,\ldots,c_4\in\mathds{R} which do not depend on \alpha. Since F \geq 0, we use \alpha F\left( \frac{M}{\alpha} \right)^k \leq \max \left\{ \alpha, \alpha F\left(\frac{M}{\alpha} \right)^3 \right\} \rightarrow 0 as \alpha \downarrow 0 for 0 \leq k \leq 3, where the convergence follows from assumption (6), to conclude t_{\min}(\tilde M) = \frac{M}{ \alpha c\left(\frac{M}{\alpha}\right) } \overset{\alpha \downarrow 0}{\rightarrow} \infty.

We verify that the arguments from the proof of Theorem 2 carry over inductively to each interval in the time-triggered setting.

Step 1: Existence of a global solution.
On each interval [t_k,t_{k+1}), the input signal u(t) = u(t_k) is constant, and the system takes the form (Proof). The existence of a maximal solution on [t_k,\tilde{\omega}_k) follows exactly as in the proof of Theorem 2. Moreover, \tilde{\omega}_k - t_k \geq t_{\min}(\tilde M), hence t_{k+1} < \tilde{\omega}_k. Concatenating the solutions on the respective intervals and using t_k \to \infty as k\to\infty, this shows the existence of a solution on [0,\infty).

Step 2: Boundedness of I.
Assume that there is a t^{\star} \in [0,\omega) such that \left\| I(t^{\star}) \right\|_{ } > \tilde{M}. Then there exists t_{\star} \in [0,t^{\star}) such that \left\| I(t_{\star}) \right\|_{ } = \tilde{M}, and \frac{d}{dt} \Vert I(t)\Vert^2_{\vert t = t_{\star}} > 0, and \left\| I(t) \right\|_{ } \leq \tilde{M} for all t \in [0,t_{\star}). Fix k \in \mathds{N} such that t_{\star} \in [t_k,t_{k+1}). Then on [0,t_{\star}], the estimate \left\| I(t) \right\|_{ } \leq \tilde{M} implies that (?), and therefore (?) hold on [0,t_{\star}]. Hence, we can use the assumption t_{k+1} - t_{k} < t_{\min}(\tilde{M}) implying (t_{k+1} - t_{k})c(\tilde{M}) < \tilde{M} to obtain the estimate

a contradiction. We can conclude that \left\| I(t) \right\|_{ } \leq \tilde{M}, for all t \in [0,\omega).

Step 3: The boundedness of u and \varphi(t) \left\| e(t) \right\|_{ } < \varepsilon carry over verbatim.