---
title: Data-Assisted Control (DAC)
url: https://www.emergentmind.com/topics/data-assisted-control-dac
type: topic
---

# Data-Assisted Control (DAC)

Data-Assisted Control (DAC) is a hybrid control architecture in which a conventional model-based nonlinear controller is continuously augmented by a data-driven correction or override when the baseline model no longer provides acceptable performance. In the aerospace and robotic formulations currently associated with the term, DAC preserves a physically grounded core dynamics model and uses data to estimate, learn, or realize the uncertain part of the closed loop, including damage effects, dissipation, actuator mismatch, contact forces, friction, and other unmodeled interactions. Recent work has recast this assistance in explicitly structured forms—most notably as a decomposition connected by a virtual port variable—so that stability arguments remain attached to the modeled subsystem while learning is confined to a lower-complexity residual map [2301.05646] [2305.12120] [2509.11688] [2509.09563] [2506.07079].

## 1. Core definition and problem setting

The 2023 aerospace formulation defines DAC as a hybrid control architecture in which a conventional, model-based nonlinear controller is continuously augmented by a data-driven correction or override when the baseline model no longer provides acceptable performance, for example after damage or under large unmodeled effects. Its stated objectives are to preserve the rigorous stability and performance guarantees of a Lyapunov-based, model-based controller over the nominal flight envelope, to estimate the true force-moment behavior in real time when unexpected dynamics appear, and to switch or convexly blend between pure model-based control and data-corrected control while retaining closed-loop stability [2301.05646].

A related 2023 flight-control variant, Sequential Data-Assisted Control (SDAC), classifies uncertainties into three types: known-predictable, known-unpredictable, and unknown. It decouples the full equations of motion into internal dynamics, where only known-predictable uncertainties appear, and external dynamics, collecting known-unpredictable and unknown effects. In that setting, a model-based nonlinear controller handles the internal dynamics and provides the desired momentum to a data-based controller responsible for the external dynamics [2305.12120].

Across later robotic formulations, the same general theme persists but is stated more abstractly. In tensor-invariant multibody control, the system is decomposed into a structurally certain, physically grounded part and an uncertain, empirical, and interaction-focused part, mediated by a virtual port variable. In port-Hamiltonian DAC, a physically meaningful observable links conservative dynamics to all actuation, dissipation, and disturbance channels, and learning is confined to the simplest part of the dynamics [2509.11688] [2509.09563]. This suggests that DAC is best understood not as end-to-end replacement of first-principles control, but as a controlled interface between known physics and learned residual behavior.

## 2. Structural decomposition and virtual-port formulations

A central development in recent DAC research is the introduction of an explicit decomposition between a modeled subsystem and an uncertain subsystem. In the tensor-invariant multibody formulation, the unreduced dynamics are written as a differential-algebraic system
$$
K\cdot \xi=b,
$$
with
$$
K=\begin{bmatrix}\mathcal M & J^\top\\ J & 0\end{bmatrix},
\qquad
\xi=\begin{bmatrix}D^I\nu\\-F_J\end{bmatrix},
\qquad
b=\begin{bmatrix}\mathcal F\\ \gamma\end{bmatrix}.
$$
This closed-form, non-recursive Newton–Euler representation keeps the internal constraint forces \(F_J\) explicit, which the paper identifies as essential for stability proofs. After rearrangement, all known tensor-mechanics terms are placed on the Left-Hand Side and the unknown interaction terms are collected on the Right-Hand Side through a generalized interaction force
$$
\tau\in\mathbb R^{6N},
$$
yielding the symbolic relation
$$
\mathrm{LHS}(\nu,D^I\nu,F_J)=\mathrm{Port}(\tau)=\mathrm{RHS}(F_B,\mathrm{Friction},\dots).
$$
The unknown mapping is then written as \(\tau=\mathcal F_\theta(\mathrm{state})\) [2509.11688].

Port-Hamiltonian DAC adopts an analogous structure. For a system
$$
\dot x=(J(x)-R(x))\nabla\mathcal H(x)+g(x)u,
$$
a virtual port variable \(\Pi\) is introduced so that
$$
\dot x-J(x)\nabla\mathcal H(x)=\Pi=-R(x)\nabla\mathcal H(x)+g(x)u.
$$
For mechanical systems, only the momentum block is affected, so \(\Pi=[\,0^\top\;\tau^\top]^T\). The papers emphasize that \(\tau\) or \(\Pi\) is not merely an algebraic convenience: it is the physically meaningful observable linking stored energy to actuators, dissipation, disturbances, or interaction forces [2509.09563] [2506.07079].

The significance of this decomposition is stated directly in several ways. The tensor-mechanics paper argues that the port cleanly separates known physics from learned uncertainty, increases explainability and interpretability, and provides a naturally ideal input for data-efficient, frame-invariant learning algorithms. The port-Hamiltonian formulations state that the structured design confines learning to the simplest part of the dynamics, enhances data efficiency, and preserves physical interpretability [2509.11688] [2509.09563]. A plausible implication is that DAC’s architectural identity lies less in a particular estimator or learner than in this decomposition principle.

## 3. Control synthesis and stability mechanisms

DAC formulations typically assign the modeled subsystem a stabilizing controller with an explicit Lyapunov or passivity proof. In the tensor-invariant multibody setting, pure velocity control uses the sliding error \(s=\nu-\nu_d\), with \(J\nu_d=0\), and the port command
$$
\tau_c=\mathcal M D^I\nu_d+\mathcal C(\nu)\nu_d-\mathcal F_{g,J}-K_d s.
$$
With the Lyapunov candidate \(V=\tfrac12 s^\top \mathcal M s\), the derivative satisfies
$$
D^I V=-s^\top K_d s\le 0,
$$
because \(Js=0\) implies \(s^\top J^\top F_J=0\), and \(D^I\mathcal M-2\mathcal C\) is skew. The same construction is extended to position-attitude-velocity tracking and to end-effector task-space control with null-space obstacle avoidance [2509.11688].

In learning-based port-Hamiltonian DAC for free-floating space manipulators, the LHS controller is a sliding-mode primary controller acting on the momentum dynamics. With
$$
s=M^{-1}(q)p-\nu,
$$
the desired port torque is
$$
\tau_c^r=M(q)\dot\nu+C'(q,p)\nu-K_d s,
$$
which yields exponentially stable error dynamics
$$
M(q)\dot s+\bigl(C'(q,p)+K_d\bigr)s=0.
$$
The same architecture adds a high-gain decentralized integrator on the RHS to drive the port-tracking error \(e_\tau=\tau_c^r-\tau_c\) rapidly to zero. The design requires the Decentralized Integral Controllability condition \(\det(\hat B_0)\neq 0\) and a strict time-scale separation in which RHS decay rates exceed the LHS sliding dynamics by a factor parameterized through \(T_d\gg 1\). The stated purpose is to ensure that subsequent learning cannot destabilize the primary dynamics [2509.09563].

The earlier GTM framework achieves continuity between model-based and data-assisted modes by introducing a decision factor \(\lambda\in[0,1]\). When \(\lambda=0\), the pure baseline controller is used; as \(\lambda\to 1\), the data-derived estimate of the generalized force-moment and its associated linear model take over. The blended controller replaces nominal quantities such as \(M,C,G,\tau_0,D\) by convex combinations of nominal and estimated values, and \(\lambda\) is updated by minimizing a short-horizon tracking cost \(J_\lambda\). The sufficient conditions stated for a stable transition are that \(\lambda(t)\) varies slowly and that the DUKF has already driven \(\hat M\to M\) so that the mismatch vanishes [2301.05646].

SDAC in flight uses robust sliding mode control for the internal dynamics. With
$$
\vartheta=\dot\eta_d-\Lambda(\eta-\eta_d),\qquad s=\dot\eta-\vartheta,
$$
the control law is
$$
\tau=J^T(\eta)\bigl(\mathcal M_\eta(\eta)\dot\vartheta+\mathcal C_\eta(\eta,v)\vartheta-\Gamma s\bigr)+u_0,
$$
and the robust term
$$
u_0=-\chi\tanh(s/\varepsilon)
$$
is introduced to eliminate chattering while preserving the inequality
$$
\dot V\le -\underline\lambda(\Gamma)\|s\|^2\le 0.
$$
The data-based outer loop then regulates momentum error through an LQR synthesized from a Koopman-identified linear model [2305.12120].

## 4. Learning, estimation, and identification layers

DAC does not prescribe a single data-driven mechanism. Instead, different formulations attach different learners or estimators to the uncertain side of the decomposition.

| Formulation | Target quantity | Method |
|---|---|---|
| GTM DAC [2301.05646] | Fixed parameters \(p\), pseudo-observed generalized force moments \(\tau_a\), and linear force-moment map | Dual Unscented Kalman Filter; Koopman least-squares estimator |
| SDAC in flight [2305.12120] | Momentum dynamics \(x_{k+1}=Ax_k+Bu_k\) for \(x\equiv \Delta L\) | Koopman-DMDc, followed by LQR recomputation |
| Tensor-invariant DAC [2509.11688] | \(\tau=\mathcal F_\theta(\nu,\mathrm{state},\tau_c)\) or \(\Delta\tau=\tau-\tau_c\) | Equivariant network, including graph-equivariant or tensor-equivariant MLP |
| DAC-pH for space manipulators [2509.09563] | \(\tau=\tilde B(x)u-\tilde D(x)\dot q+d(t)\) | Physics-informed neural network trained online with ADAM |
| Generalized DAC-pH [2506.07079] | RHS policy mapping \((x,\Pi_c)\mapsto u\) | Reinforcement learning, including Soft-Actor-Critic |

In the GTM framework, the DUKF estimates the augmented state \(x_k=[v;\tau_a;p]\), where \(\tau_a=[\tau;\zeta_1;\zeta_2]\) and \(\tau\) follows a 3rd-order Gauss-Markov process. The subsequent Koopman estimator constructs
$$
\hat{\mathcal P}=\mathcal T\mathcal Y^\top(\mathcal Y\mathcal Y^\top)^{-1}
$$
for the mapping \([B\;D\;\tau_0]\). The paper explicitly notes that richer observables could also be used, including products, higher-order monomials, extended DMD, or SINDy [2301.05646].

In SDAC, the external dynamics are expressed around a trim through the momentum deviation \(\Delta L\), then discretized and identified from data using DMDc under the observable \(g(x)=x=\Delta L\). Every data window \(P_w\), the pair \((A,B)\) is re-estimated and the LQR gain is recomputed, provided the controllability matrix has full rank [2305.12120].

Tensor-invariant DAC makes frame invariance an explicit design principle. The learner receives tensor inputs such as twists, forces, joint states, and \(\tau_c\), and, when implemented with \(E(3)\)-equivariant layers or another equivariant architecture, commutes with rotations and translations of the entire mechanism. The stated consequence is that training data collected in one orientation or workspace generalizes to new frames without re-learning gravitational or inertial couplings [2509.11688].

The free-floating space-manipulator formulation uses a PINN with two subnetworks for \(\hat B(x)\) and \(\hat D(x)\), using monomials and trigonometric functions of \(x\) as input features and enforcing positive dissipation through a \(\softplus\)-parameterized diagonal output for \(\hat D(x)\). The loss \(\ell=\|\breve\tau-\hat\tau\|^2\) is minimized online with ADAM, while the surrounding integrator and switching term are designed to keep learning transients bounded [2509.09563].

## 5. Applications and reported empirical results

Several papers provide simulation studies intended to validate distinct DAC variants rather than a single standardized benchmark.

| Application | Setup | Reported results |
|---|---|---|
| NASA Generic Transport Model [2301.05646] | 5.5% scale GTM; damage at \(t=10\) s; added high-frequency yaw-moment term at \(t=30\) s | Pure model-based control yields velocity error jump to \(\sim 0.5\) ft/s; under DAC, velocity error returns below \(0.05\) ft/s within 5 s; Koopman estimate error remains \(\sim 10^{-3}\); \(\lambda(t)\) rises from 0 to 1 over \(\sim 3\) s after damage |
| SDAC in flight [2305.12120] | GTM cruise trim; uncertainty in \(D,B\) at \(t=20\) s; identification window \(P_w=10\) s | \(\|(A,B)\|\) estimation error drops below 5% after one window; RMS attitude error reduced by \(\sim 40\%\); RMS velocity error reduced by \(\sim 35\%\); peak momentum overshoot cut by \(\sim 50\%\) |
| Tensor-invariant multibody DAC [2509.11688] | 5-body serial chain; open-loop and end-effector tracking tests | Total linear/angular momentum error \(<10^{-6}\); constraint violation \(\|J\nu\|<10^{-8}\); max position error \(<2\) mm; max orientation error \(<0.5^\circ\); settling time \(<0.2\) s post-disturbance; equivariant network converged within 500 samples vs. 5,000 for a non-equivariant baseline |
| Free-floating space manipulator DAC-pH [2509.09563] | 2-DOF planar manipulator with link masses \((2,1,1)\) kg and lengths \((0.35,0.35)\) m | By end of phase 2, relative \(L_2\) estimation errors dropped to 63% for \(\tilde B\) and 30% for \(\tilde D\); DAC vs. model-based gives 77.5% reduction in attitude tracking error, 16% increase in control effort, and full rejection of high-impact disturbance |
| Pendulum DAC-pH [2506.07079] | \(T_s=0.05\) s, \(m=1\) kg, \(l=1\) m, \(c=0.1\) N·m·s | 10 random-IC trials show stable convergence \(q\to 0\), \(p\to 0\); moving-average reward declines monotonically over \(\sim 50\) episodes; without integrator \(q_{stst}\approx 0.05\) rad; with integrator \(q\to 0\) exactly and \(\tau\to 0\) |

Taken together, these case studies support a consistent empirical picture. DAC is used where a purely model-based controller degrades under damage, contact, actuation uncertainty, or unmodeled disturbances, while the addition of a structured estimator or learner restores tracking, reduces residual error, or reduces sample complexity. The tensor-invariant results add a specific geometric claim: when the known physics is factored out and the learner respects symmetry, sample complexity can drop by an order of magnitude on the reported benchmark [2509.11688].

## 6. Terminological scope and common ambiguities

The acronym “DAC” is not used uniformly across adjacent literatures. In adaptive online non-stochastic control, “DAC” denotes a disturbance action controller with policy
$$
u_t=Kx_t+\sum_{i=1}^H M_{t,i}w_{t-i},
$$
embedded in an AdaFTRL-C update for the matrices \(M_t\). That work proves a data-adaptive policy-regret bound
$$
\mathrm{Regret}_T\le \frac{2}{\delta}\sqrt{2\kappa_M^2(\delta^2+2lw p\kappa_B)}\cdot \sqrt{\sum_{t=1}^T g_t},
$$
and explicitly remarks that the matrices \(M_{t,i}\) adaptively “learn” a model of how past disturbances affect future costs, complementing the robust baseline \(K\) [2310.02261].

In a different community, DAC denotes Dynamic Algorithm Configuration. There, the problem is formulated as an MDP \(M=(S,A,P,R)\) with a factored action space \(A=A_1\times\cdots\times A_M\), and the CANDID benchmark studies Coupled Action Dimensions with Importance Differences through sequential policies of the form
$$
\pi(a|s)=\prod_{m=1}^M \pi_m(a^m\mid s,a^{1:m-1}).
$$
The reported result is that sequential policies outperform independent learning of factorized policies in those action spaces [2407.05789].

This suggests that “Data-Assisted Control” should be disambiguated carefully from other established expansions of the same acronym. Within robotics, aerospace, and port-Hamiltonian control, DAC refers to hybrid physics-plus-data architectures centered on a modeled subsystem, an uncertain subsystem, and an interface variable such as \(\tau\), \(\Pi\), \(L_d\), or \(\lambda\). In online control and algorithm-configuration literatures, the same initials may denote different formal objects and different problem classes.

Source: https://www.emergentmind.com/topics/data-assisted-control-dac