Maximum Hands-Off Hybrid Control Sequences

• Maximum hands-off hybrid control sequences are strategies that minimize active control intervals in hybrid systems, reducing energy consumption and actuator wear.
• The methodology integrates convex relaxations, such as Elastic-Net and CLOT norms, with combinatorial and graph-based approaches to obtain sparse yet smooth control actions.
• Practical applications include electric vehicles and industrial automation where reducing actuator interventions optimizes system performance and longevity.

Maximum hands-off hybrid control sequences are control strategies for dynamical systems, particularly switched or hybrid systems, that minimize the use of actuators by maximizing the intervals during which both the continuous input and switching actions are held at zero or remain unchanged. These sequences are central to energy-saving control, actuator preservation, and systems where discontinuous actuation is either undesirable or physically costly. This entry synthesizes results from convex and combinatorial approaches, state-space abstraction, and invariant-set methods, as developed in the contemporary literature (Challapalli et al., 2017, Challapalli et al., 2016, Sperilă et al., 14 Mar 2025, U et al., 25 Jan 2026).

1. Problem Formulation: Hands-Off Control in Hybrid and Switched Systems

The maximum hands-off control paradigm seeks control sequences that transfer a system’s state from a given initial condition $\xi$ to a target (often the origin), within a prescribed time, while minimizing the nonzero support of both continuous actuators and switching operations. For linear time-invariant plants, the classical hands-off problem restricts itself to continuous controls $u(t)\in\mathbb{R}$, while hybrid and switched systems extend this to discrete switching sequences $\nu(t)\in\{1,\dots,N\}$. In discrete time, the switched system is represented as

$$x(t+1) = A_{\nu(t)} x(t) + b_{\nu(t)} \mu(t), \quad x(0)=\xi,$$

with admissibility constraints on $(\nu(t),\nu(t+1))\in E(N)$ and $\mu(t)\in U$ (U et al., 25 Jan 2026). The goal is to minimize the cumulative number of discrete switches ($\|\Delta(\nu_T)\|_0$) and the number of nonzero continuous inputs ($\|\mu_T\|_0$), steering $x$ to the origin in $T$ steps. In continuous-time, the maximum hands-off objective with bounded amplitude is typically formalized as minimizing the $L^0$ "norm" of the control input, i.e.,

$$\min_{u \in \mathcal{U}} \|u\|_0 = \min_{u\in\mathcal{U}} \int_0^T \phi_0(u(t))\,dt,$$

subject to system dynamics $\dot x(t) = A x(t) + B u(t)$ and $|u(t)|\leq U_\text{max}$ (Challapalli et al., 2017, Challapalli et al., 2016).

2. Structure of Maximum Hands-Off Sequences: Bang–Off–Bang and Sparsity

For controllable systems with nonsingular $A$, the $L^0$-optimal (maximum hands-off) solution has a bang-off-bang structure: $u^*(t)\in\{\pm U_\text{max}, 0\}$ almost everywhere. The optimal support is the sparsest among all feasible controls. In hybrid settings, sparsity is measured in both the switching and actuation dimensions. The sparsest hybrid sequence minimizes

$\|\Delta(\nu_T)\|_0 + \|\mu_T\|_0$

over all admissible pairs steering the system to target (U et al., 25 Jan 2026). These combinatorial objectives are nonconvex, and exact solutions are computationally hard except in restricted cases or via abstraction.

3. Convex Relaxations: Elastic-Net, CLOT, and State Invariance Formulations

Convex relaxations provide practical means to compute approximate maximum hands-off controls while ensuring continuity or robustness:

• Elastic-Net (EN) Control: Augments the LASSO ($L^1$) sparsity term with an $L^2$-squared term:

$\min_{u\in\mathcal{U}} \|u\|_1 + \lambda \|u\|_2^2$

which promotes smooth but less sparse controls (Challapalli et al., 2017, Challapalli et al., 2016).

• CLOT Norm Control: The CLOT approach combines $L^1$ and non-squared $L^2$ penalties:

$\min_{u\in\mathcal{U}} \|u\|_1 + \lambda\|u\|_2$

yielding controllers that are continuous (in the limit $h\to0$) yet have sparser support than EN controls. The resulting CLOT-optimal control has the property that the difference between adjacent discretized values satisfies $|u_k^*-u_{k+1}^*|=O(\sqrt{h})$ (Challapalli et al., 2017). This continuity is provable via KKT conditions and subdifferential analysis.

• Set-based Invariance Approaches: For discrete-time systems with explicit switching, hands-off regions in the state space are defined, and actuation is applied only to bring the state into a designated invariant set $\Omega_I$, after which the control can be switched off ($\sigma=0$). Closed-loop feedback is engaged as required to maintain constraint satisfaction, yielding a two-mode hybrid controller with maximal open-loop intervals (Sperilă et al., 14 Mar 2025).

4. Combinatorial and Graph-Based Synthesis Methods

For switched systems, explicit enumeration and combinatorial optimization algorithms have been developed:

• State-space Partitioning: Partition $X$ into cells $\{X_0,X_1,...,X_n\}$, mapping each to vertices in a directed, labeled multigraph $G$.
• Transition Graph Construction: Edges represent transitions via the discrete mode and zero or nonzero continuous actions. Each walk in $G$ encodes a candidate hybrid sequence.
• Weight Assignment and Walk Selection: The weight of a walk corresponds to the total number of switches and continuous inputs used. The minimum-weight walk from the initial-state cell to the target ($X_0=\{0\}$) corresponds to an admissible maximum hands-off sequence (U et al., 25 Jan 2026).

This method provides certifiably optimal hands-off sequences under abstraction, with existence conditions guaranteed if the transition-closure property holds for the partition. For classes such as diagonal or anti-diagonal subsystems with appropriately structured input maps, explicit partitions ensuring transition closure are constructible (U et al., 25 Jan 2026).

5. Tradeoffs: Sparsity, Continuity, and Performance

Table: Sparsity Densities under Different Relaxations (For a 4th-order integrator, $T=20$, $\lambda=0.1$)

Method	Density ($u(t)\neq0$ fraction)
LASSO	0.17
EN	0.38
CLOT	0.27

CLOT controllers provide significantly longer intervals of zero control compared to EN, while maintaining continuity, unlike LASSO which is discontinuous (Challapalli et al., 2017). Parameter selection for $\alpha$ in the CLOT cost enables direct tuning of the tradeoff: larger $\alpha$ yields greater sparsity with sharper transitions, while smaller values favor smoother but less hands-off control (Challapalli et al., 2016).

6. Extensions: State Constraints, Feedback, and System Classes

State constraints $\|x(t)\|_2\leq\theta$ can be incorporated directly into the CLOT- or EN-based discretized convex formulations, leading to second-order cone programs solvable by standard convex optimization tools (Challapalli et al., 2017).

For feedback-augmented hands-off control, invariant set-based methods systematically exploit state feedback to drive the system into a hands-off region and then maintain constraint satisfaction in open-loop. This yields provable maximum dwell times for hands-on phases and high hands-off ratios, even under disturbance and noise, provided the sets $\Omega_I,\Omega_O$ are chosen appropriately via convex LMI procedures (Sperilă et al., 14 Mar 2025).

Hybrid extensions to underactuated, piecewise-affine, or networked systems are possible by integrating CLOT-like costs into mixed-integer convex programming or by extending the invariant-set approach to nonlinear dynamics (Challapalli et al., 2016, Sperilă et al., 14 Mar 2025).

7. Applications and Practical Considerations

Maximum hands-off hybrid control finds direct application in domains such as electric vehicles, energy-efficient actuators, coasting trains, and practical hybrid plants requiring both minimal actuator engagement and smooth transitions. Numerically, CLOT-based methods are tractable for problem sizes (e.g., $N\sim10^3$), while graph-based combinatorial methods are suited to small-to-moderate dimensional switched systems with explicitly tractable abstraction. Robustness to measurement noise and model perturbation can be systematically addressed in the invariant-set-based approach.

Convex relaxations yield controls that interpolate between discontinuous bang-off-bang and smooth dense strategies, while graph-theoretic approaches provide certifiable optima in settings admitting finite cell decomposition (Challapalli et al., 2017, U et al., 25 Jan 2026). The tradeoffs among sparsity, continuity, and robustness are explicit, tunable, and central to the synthesis of practically useful hybrid hands-off control sequences.