2022 IEEE 61st Conference on Decision and Control (CDC)

December 6-9, 2022. Cancún, Mexico

State and Input Constrained Model Reference Adaptive Control

Poulomee Ghosh and Shubhendu Bhasin

Abstract— Satisfaction of state and input constraints is one approach [11], [12] etc. Most of these techniques involve
2022 IEEE 61st Conference on Decision and Control (CDC) | 978-1-6654-6761-2/22/$31.00 ©2022 IEEE | DOI: 10.1109/CDC51059.2022.9992849

of the most critical requirements in control engineering appli- solving an optimization problem at every time instant which
cations. In classical model reference adaptive control (MRAC) can be computationally expensive. To deal with these prob-
formulation, although the states and the input remain bounded,
the bound is neither user-defined nor known a-priori. In lems, various tools like Barrier Lyapunov Function (BLF)
this paper, an MRAC is developed for multivariable linear [13], Control Barrier Function(CBF) [14] etc. have been
time-invariant (LTI) plant with user-defined state and input utilized to guarantee safety. The safety certificate of CBF was
constraints using a simple saturated control design coupled introduced in [14] where CBF and control Lyapunov function
with a barrier Lyapunov function (BLF). Without any restric- (CLF) are unified through quadratic programs (CBF-CLF-QP
tive assumptions that may limit practical implementation, the
proposed controller guarantees that both the plant state and the approach) to ensure safety in terms of forward invariance
control input remain within a user-defined safe set for all time of set [15], [16]. For CBF-CLF-QP approach, although the
while simultaneously ensuring that the plant state trajectory controller guarantees safety, the system trajectory doesn’t
tracks the reference model trajectory. The controller ensures essentially converge to the origin. A generalized approach of
that all the closed-loop signals remain bounded and the trajec- BLF was presented in [13], [17] to satisfy safety constraints
tory tracking error converges to zero asymptotically. Simulation
results validate the efficacy of the proposed constrained MRAC for output feedback control systems. In [18], BLF is used
in terms of better tracking performance and limited control with model reference adaptive controller for constraining
effort compared to the standard MRAC algorithm. trajectory tracking error and adaptive gains within user-
defined sets where the estimate of the controller parameters
I. I NTRODUCTION is assumed to be very close to the actual value. An alter-
Modern control applications are characterized by physical, native approach to ensure safety is the state transformation
safety and energy limitations that can often be translated technique using BLF. An adaptive tracking controller is
into state and input constraints on the plant dynamics. With- developed in transformed state space using BLF to guarantee
out accounting for their effect during control design, these performance bound in [19]. Although BLF-based controllers
constraints are often met using an ad-hoc approach. While ensure that user-defined state constraints are met, they usu-
conventional adaptive control techniques manage to control ally result in large control effort when the states approach
systems under parametric uncertainty, they fail to adhere to the boundary of the constrained region, often violating the
user-defined constraints. This challenge is exacerbated for actuator’s operating limits.
safety critical systems where safety is of utmost importance. Constraining the input to account for actuator saturation lim-
A particular versatile class of adaptive controllers is model its is another issue of practical concern that has been tackled
reference adaptive controllers (MRAC) that aim to control extensively in literature, especially using various saturated
systems with parametric and matched uncertainties by track- functions e.g. hyperbolic tangent, sigmoid etc. [20]–[24]. An
ing a user-defined stable reference model system [1]–[4]. adaptive controller is developed in [25], [26] for a single-
Although the control law ensures asymptotic tracking, the input single-output (SISO) LTI plant in the presence of input
bound on the tracking error is typically not known a-priori. constraints. In [27], an adaptive tracking control method has
Moreover, classical MRAC does not allow for user-defined been investigated for MIMO nonlinear systems where an
constraints on the state and the input. Large magnitude of auxiliary design system was introduced to deal with input
control effort might exceed actuator’s saturation limit and constraints.
cause deterioration of the process. Hence, constraining the All the aforementioned approaches either involve state or
plant states and input within known user-defined bounds input constraints. Few control approaches exist that constrain
while meeting satisfactory performance objectives is a prob- both state and input for uncertain systems. MPC [28]–[30]
lem of practical interest. is a popular control approach where both state and input
Several control approaches have been proposed that either constraints can be included in the optimization routine, albeit
partially or fully address these challenges. The state con- at the cost of computational complexity. Further, limited
strained tracking control problem has been dealt with using or imperfect model knowledge often leads to conservative
model predictive control [5], [6], optimal control theory results. A recent work [31] develops an MRAC law that
[7], [8], invariant set theory [9], [10], reference governor places user-defined bounds on state and input. The result,
however, is achieved by developing an auxiliary reference
Poulomee Ghosh and Shubhendu Bhasin are with Department of model that complicates the analysis and design of adaptive
Electrical Engineering, Indian Institute of Technology Delhi, New
Delhi, India. (Email: [email protected], laws.
[email protected]) The main contribution of the work is the development

978-1-6654-6761-2/22/$31.00 ©2022 IEEE 68

Authorized licensed use limited to: University Duisburg Essen. Downloaded on June 12,2023 at 13:45:51 UTC from IEEE Xplore. Restrictions apply.
of MRAC for multivariable LTI systems while satisfying Assumption 1: For any β > 0, there exist positive con-
safety constraints on both the state and the input. To this stants α1 , α2 ∈ R such that
end, a saturated controller with BLF-based adaptation laws
are designed. Inspired by [25], the design of the saturated ∥xr (t)∥ ≤ α1 < β (3)
controller is intuitive and simple, while the classical MRAC ∥ẋr (t)∥ ≤ α2 (4)
adaptation laws are modified based on BLF to ensure that Assumption 2: There exists a feasible control policy u(t)
the states and the input always lie in a user-defined safe that satisfies both the input and state constraints for all time.
region. Closed-loop signals are guaranteed to be bounded Control Objective: The control objective is to design
and the trajectory tracking error can be proved to converge an input u(t), such that x(t) tracks xr (t) i.e. e(t) ≜
to zero asymptotically. The adaptation rate being unchanged, x(t) − xr (t) → 0 as t → ∞ while both the input and
the proposed controller shows better tracking performance the state constraints are satisfied. Using Assumption 1, the
than the classical MRAC while simultaneously guaranteeing state constraints can be transformed to the constraint on the
the safety constraints. tracking error: ∥e(t)∥ < kb , ∀t ≥ 0, where kb ∈ R is a
This paper is organized as follows. Section II presents positive constant given by kb = β −α1 , i.e. ∥e(t)∥ < kb =⇒
the problem formulation and preliminaries of standard ∥x(t)∥ < β.
MRAC framework, section III elucidates proposed BLF-
based methodology which satisfies state and input con-
straints. To justify the preeminence of the proposed con- B. Classical MRAC
troller, simulation results and comparative analysis are shown Consider the classical certainty equivalence adaptive con-
in section IV while section V comprises of conclusion and trol law
future works.
u = K̂x x + K̂r r (5)
II. P ROBLEM F ORMULATION
where K̂x (t) ∈ Rm×n and K̂r (t) ∈ Rm×m are estimates
Throughout this paper R denotes the set of real numbers, of the controller parameters Kx and Kr , respectively which
Rp×q denotes set of p × q real matrices, the identity matrix satisfy the matching conditions given in Assumption 2.
in Rp×p is denoted by Ip and ∥.∥ represents the Euclidian Assumption 3: There exists controller parameters Kx ∈
vector norm and corresponding equi-induced matrix norm. Rm×n and Kr ∈ Rm×m such that the following matching
conditions are satisfied.
A. Problem Statement
Consider a linear time-invariant system A + BKx = Ar
BKr = Br (6)
ẋ(t) = Ax(t) + Bu(t) (1) Using (5) and (6) , the closed-loop error dynamics can be
obtained as
where x(t) ∈ Rn denotes the system state, u(t) ∈ Rm
denotes the control input, A ∈ Rn×n is the unknown system ė = Ar e + B K̃x x + B K̃r r (7)
matrix, and B ∈ Rn×m is the input matrix, assumed to
be full rank and known. The pair (A,B) is assumed to be where K̃x (t) ≜ K̂x (t) − Kx ∈ Rm×n and K̃r (t) ≜ K̂r (t) −
stabilizable. Kr ∈ Rm×m denote the parameter estimation errors.
A reference model is considered as The classical adaptive update laws are given as [1]

ẋr (t) = Ar xr (t) + Br r(t) (2) ˙

K̂x = −Γx B T P exT
where xr (t) ∈ Rn is the state of the reference model, ˙
K̂r = −Γr B T P erT (8)
r(t) ∈ Rm is a bounded piecewise continuous reference
input, Ar ∈ Rn×n , Br ∈ Rn×m are known. It is assumed where, Γx ∈ Rm×m and Γr ∈ Rm×m are positive definite
that Ar is Hurwitz i.e. for every Q = QT > 0, there exists adaptation gain matrices. using Lyapunov analysis, we can
P = P T > 0 such that ATr P + P Ar + Q = 0. prove that all the closed-loop signals remain bounded and the
trajectory tracking error converges to zero asymptotically [1],
State Constraint: For any positive constant β, the [2].
system states should remain within a user defined safe set Note that, actuator limits and state constraints are typically
given by Ωx := {x ∈ Rn : ∥x∥ < β}. not considered in the classical MRAC design, rather they are
often imposed in an ad-hoc manner during implementation.
Input Constraint: Magnitude of the control input The focus of the work is to consider both input and state
should remain bounded in a safe set given by constraints in the MRAC design procedure and rigorously
Ωu := {u ∈ Rm : ∥u∥ ≤ umax }, where umax is a show boundedness of closed-loop signals and provide stabil-
user-defined positive constant. ity guarantees.

69
Authorized licensed use limited to: University Duisburg Essen. Downloaded on June 12,2023 at 13:45:51 UTC from IEEE Xplore. Restrictions apply.
 III. P ROPOSED M ETHODOLOGY V1 (e) : Ωe → R, defined in an open region containing the
origin such that
A. Input Constraint Satisfaction Using Saturated Control
Design
Consider the linear time-invariant plant given in (1) and V1 (e) → ∞ ∥e∥ → kb (15)
reference model given in (2). Consider an auxiliary control
input v(t) ∈ Rm as The candidate Lyapunov function can be written as V (µ) =
V1 (e)+V2 (ξ). Given Assumption 4, if the following inequal-
v(t) = K̂x x + K̂r r (9)
ity holds
where v(t) ≜ [v1 (t), . . . , vm (t)]T . Inspired by [25], the
saturated feedback controller is designed for the vector case ∂V
V̇ = f ≤0 (16)
as ∂µ
(
vi (t) if |vi (t)| ≤ u√max
m
ui (t) = umax u√ , i = 1, . . . , m then e(t) ∈ Ωe ∀t.
√
m
sgn(v i (t)) if |v i (t)| > max
m Proof: For the proof of Lemma 1, see [13].
(10)
To ensure constraint satisfaction on the trajectory tracking
where u(t) ≜ [u1 (t), . . . , um (t)] . The closed-loop error error, consider a BLF
T

dynamics is given as

ė = Ar e + B K̃x x + B K̃r r + B∆u (11) ′

1 k2
V1 (e) ≜ log ′ 2 b T (17)
where ∆u(t) ∈ Rm is defined as the difference between 2 kb − e P e
actual control input and auxiliary control input : ∆u(t) ≜
u(t) − v(t). To mitigate the effect of the disturbance term ′
n T ′2 ′

B∆u(t), consider an auxiliary error signal e1 (t) with the p on the set ΩTe : {e ∈ R ′2 : e P e < kb }, where kb =
defined
following dynamics kb λmin {P }. If e P e → kb , i.e. when the constrained
state e(t) approaches the boundary of the safe set, the BLF
ė1 = Ar e1 + K1 (t)∆u e1 (t0 ) = 0 (12) V1 (e) → ∞, guaranteeing the safety of the system. The
unconstrained states involve continuously differentiable and
where K1 (t) ∈ Rn×m is time-varying controller parameter. positive definite quadratic functions.
Let ed (t) be the difference between the actual and auxiliary
Consider the candidate Lyapunov function V (µ) : ξ×R →
error signals: ed (t) ≜ e(t) − e1 (t). The error dynamics of
R as,
ed (t) is given by
′2
ėd = Ar ed + B K̃x x + B K̃r r + Kd ∆u (13) 1

kb
V (µ) = log ′2 + eTd P ed + tr(K̃xT Γ−1
x K̃x )
where Kd (t) ≜ B − K1 (t). 2 kb − eT P e

+ tr(K̃rT Γ−1
r K̃r ) + tr(KdT Γ−1
d Kd ) + tr(K1T Γ−1
1 K1 )

B. State Constraint Satisfaction using BLF (18)

To ensure state constraint satisfaction, a BLF is introduced
[13]. where Γd ∈ Rn×n and Γ1 ∈ Rn×n are positive-definite
Assumption 4: The initial conditions x(0) are chosen matrices. Taking the time-derivative of V along the system
such that the initial trajectory tracking error is bounded as trajectory
∥e(0)∥ < kb .
Lemma 1: For any positive constant kb , let Ωe := {e ∈ 1

Rn : ∥e∥ < kb } ⊂ Rn and Ψ := RN × ζ ⊂ RN +n be open V̇ = ′2 eT P (Ar e + B K̃x x + B K̃r r + B∆u)
sets. Consider the system dynamics given by 2(kb − eT P e)

µ̇ = f (t, µ) (14) + (Ar e + B K̃x x + B K̃r r + B∆u)T P e

1 T
µ := [eT , ξ T ]T ∈ Ψ, where ξ := [K̃x , K̃r , Kd , K1 ] is the + ed P (Ar ed + B K̃x x + B K̃r r + Kd ∆u)
augmentation of the unconstrained states and the function f : 2

R+ × Ψ → RN +n is measurable for each fixed µ and locally + (Ar ed + B K̃x x + B K̃r r + Kd ∆u)T P ed
Lipschitz in e, piecewise continuous and locally integrable
on t. Suppose, there exists positive definite, decrescent, ˙ T −1 ˙
+ tr(K̃xT Γ−1
x K̂x ) + tr(K̃r Γr K̂r )
quadratic candidate Lyapunov function V2 (ξ) : RN → R and
continuously differentiable, positive definite, scalar function + tr(KdT Γ−1 T −1
d K̇d ) + tr(K1 Γ1 K̇1 ) (19)

70
Authorized licensed use limited to: University Duisburg Essen. Downloaded on June 12,2023 at 13:45:51 UTC from IEEE Xplore. Restrictions apply.
Substituting B = Kd + K1 in (19), i.e. the trajectory tracking error will be constrained within
 the user-defined safe set : e(t) ∈ Ωe ∀t ≥ 0.
1
V̇ = ′2 eT (ATr P + P Ar )e + eT BP K̃x x Further, since x(t) = e(t) + xr (t) and the reference model
2(kb − eT P e) states and the trajectory tracking error is bounded, i.e.
+ eT BP K̃r r + eT P (Kd + K1 )T ∆u + xT K̃xT B T P e ∥xr (t)∥ ≤ α1 , ∥e(t)∥ < kb , it can be easily shown that
the proposed controller guarantees the plants states to be
 
1
+ rT K̃rT B T P e + ∆uT (Kd + K1 )T P e + eTd (ATr P bounded within the user defined safe set
2
+ P Ar )ed + eTd P B K̃x x + eTd P B K̃r r + eTd P Kd ∆u
 ∥x(t)∥ < kb + α1 < β ∀t ≥ 0 (27)
T T T T T T T T
+ x K̃x B P ed + r K̃r B P ed + ∆u Kd ed
˙ T −1 ˙ T −1 Thus the state constraint gets satisfied, i.e. x(t) ∈ Ωx for all
+ tr(K̃xT Γ−1
x K̂x ) + tr(K̃r Γr K̂r ) + tr(Kd Γd K̇d ) t ≥ 0.
+ tr(K1T Γ−1
1 K̇1 ) (20)
Adaptive update laws are defined as (ii) The control effort of the proposed controller
u(t) = [u1 (t), . . . , um (t)]T and ∥u(t)∥ =
Γx B T P exT
  p
2 2
u1 (t) + u2 (t) + . . . + u2m (t). For constraining the
˙ T T
K̂x = − ′2 + Γx B P ed x
k − eT P e control input two cases have been considered.
 b T Case 1: ∥vi (t)∥ ≤ u√max
Γr B P erT

˙ T T m
K̂r = − ′2 + Γr B P ed r For this case, ui (t) = vi (t) and ∆u(t) = 0. So, |ui | ≤ u√max
kb − eT P e m
which implies ∥u∥ < umax
Γd P e∆uT
 
K̇d = − ′2 + Γ d P ed ∆uT Case 2: ∥vi (t)∥ > u√max m
kb − eT P e For this case, ui (t) = u√max sgn(vi (t)) which proves
m
Γ1 P e∆uT ∥u∥ < umax .
K̇1 = − ′2 (21)
kb − eT P e
which yields (iii) Since the closed loop tracking error as well as
the controller parameter estimation errors remain bounded
eT Qe
 
1 and Kx (t) and Kr (t) are constants, it can be concluded
V̇ = − ′2 + eTd Qed ≤0 (22)
2 kb − eT P e that the estimated parameters are also bounded i.e.
K̂x (t), K̂r (t) ∈ L∞ followed by ensuring the plant state
which is a negative semi-definite function.
x(t) and control input u(t) to be bounded for all time
Theorem 1: Consider the linear time-invariant plant (1)
instances. Thus the proposed controller guarantees all the
and reference model (2). Given Assumptions 1-3, the pro-
the closed loop signals to be bounded.
posed controller (9), (10) and the adaptive laws (21) ensure
that the following properties are satisfied.
(iv) Since V (µ) > 0 and V̇ (µ) is negative semi-definite
(i) The plant states remain within the user-defined safe set
(22), it can be shown that e, K̃x , K̃r , Kd , K1 ∈ L∞ ,
given by Ωx := {x ∈ Rn : ∥x(t)∥ < β}.
x(t) ∈ L∞ , and K̂x , K̂r ∈ L∞ . Further, from (22) it can
(ii) The control effort is bounded within a user-defined safe
be shown that e(t) ∈ L2 and from (11) it can be inferred
set given by Ωu := {u ∈ Rm : ∥u∥ ≤ umax }.
that ė(t) ∈ L∞ . Therefore, e(t) is uniformly continuous.
(iii) All the closed loop signals remain bounded.
Consequently, using Barbalat’s Lemma [32], it can be proved
(iv) The trajectory tracking error converges to zero asymp-
that e(t) converges to zero asymptotically as t → ∞.
totically i.e. e(t) → 0 as t → ∞.
Proof: (i) V (µ) in (18) is positive definite and V̇ (µ) ≤
0 from (22), which implies that V (µ(t)) ≤ V (µ(0)) ∀t ≥ 0.
′ IV. S IMULATION R ESULTS
As V (µ) is defined in the region Ωe := {[eT , ξ T ] ∈ Ψ :
′2
eT P e ≤ kb }, it can be inferred from Lemma 1 that
′2
To demonstrate the efficacy of the proposed algorithm, a
eT P e < k b (23) multivariable LTI plant and reference model are considered.
=⇒ eT P e < λmin {P }kb2 (24)
Now, for any positive-definite matrix P , 
−0.322 0.064 0.0364 −0.9917 0.0003 0.0008 0
 
0 0

   
0 0 1 0.0037 0 0 0  0 0 
eT P e ≥ λmin {P }∥e∥2 (25)
 
   
   
−30.6492 0 −3.6784 0.6646 −0.7333 0.1315 0   0 0 
   
Given Assumption 3, from (24) and (25) it can be proved A= 8.5396 0 −0.0254−0.4764−0.0319−0.0620 0


B= 0 0 






that 


0 30 0 10 20.2 0 0



10.1 0 




 −1 0 10 0 0 10.25 0   0 −4.25
∥e∥ < kb ∀t ≥ 0
   
(26) 0 0 0 12.2958 0 0 −1 0 0

71
Authorized licensed use limited to: University Duisburg Essen. Downloaded on June 12,2023 at 13:45:51 UTC from IEEE Xplore. Restrictions apply.
    
−0.322 0.064 0.0364 −0.9917 0.0003 0.0008 0 0 0
   

 0 0 1 0.0037 0 0 0 

 0 0 
 
   
−30.6492 0 −3.6784 0.6646 −0.7333 0.1315 0   0 0 
   
Ar = 8.5396 0 −0.0254−0.4764−0.0319−0.0620 0 Br = 0 0 
   
   
0 0 0 0 −20.2 0 0 20.20 
   
 
   

 0 0 0 0 0 −20.2 0 

 020.2
 
0 0 0 12.2958 0 0 −1 0 0

Fig. 2. Control input using proposed controller (21) and classical MRAC
The reference signal is considered as: r(t) = (8).
[exp(−t/10); exp(−t/20)].
  The other
 parameters
 are
5 0 5 0
chosen as: Γx = , Γr = , Γ1 = In×n ,
0 5 0 5
Γd = In×n , umax = 2.5, β = 2, α1 = 1.5 and kb = 0.5.
The desired controller must satisfy the input constraint
∥u∥ ≤ 2.5, while ensuring that plant states within
user-defined bound i.e. ∥x∥ < 2. Given Assumption 1,
∥xr ∥ ≤ 1.5, the state constraint is equivalent to satisfying
the constraint in the error i.e. ∥e∥ < 0.5. It is assumed that
the initial plant states remain within the user-defined safe
set i.e. ∥x(0)∥ < 2.
To gauge the safety and performance of the proposed control
law, we compare it with the classical MRAC controller Fig. 3. Tracking performance of the plant using proposed controller (21)
and classical MRAC (8).
(in (5) and (8)). The reference signal is considered as:
r(t) = [exp(−t/10);
 exp(−t/20)].
 The adaptation
 gains are
25 0 25 0
chosen as: Γx = , Γr = . It is seen that increasing the adaptation gain leads to
0 25 0 25
Note that, adaptation gains are tuned to achieve better better tracking performance, in general for both the classical
tracking performance for both proposed controller and and the proposed controller, although the response becomes
classical MRAC. more oscillatory. The improved tracking performance of the
classical MRAC is achieved at the cost of greater control
effort leading to violation of the input constraints. Further,
in the classical MRAC case, the high frequency oscillation in
the control input may even violate the actuation rate limits.
On the other hand, the state and the input constraints are
never violated in case of the proposed controller.

V. C ONCLUSION
In this paper, a novel MRAC architecture is proposed
for multivariable LTI systems by strategically combining
BLF with a saturated controller which guarantees both the
plant state and the control input remain bounded within
Fig. 1. Comparative analysis of trajectory tracking error using the proposed user-defined safe sets. The proposed controller also ensures
control law (21) and classical MRAC law (8).
that the trajectory tracking error asymptotically converges to
zero and the closed-loop signals remain bounded. Simulation
Fig.1 shows the trajectory tracking error using the pro- studies validate the efficacy of the proposed control law com-
posed method where the user-defined constraint is satisfied paring to classical MRAC. Extending the work to uncertain
while in conventional MRAC case, the norm of the trajectory nonlinear systems and exploring robustness properties is an
tracking error goes beyond the safe region. Furthermore, the important area of future research.
proposed controller ensures that the tracking error converges
to zero as time tends to infinity and the rate of convergence is R EFERENCES
higher than the classical MRAC. The proposed control archi- [1] K. S. Narendra and A. M. Annaswamy, Stable adaptive systems.
tecture bounds the control effort in user-defined constrained Courier Corporation, 2012.
region (Fig. 2) while for the conventional MRAC the bound [2] P. Ioannou and B. Fidan, Adaptive control tutorial. SIAM, 2006.
[3] B. Shackcloth and R. B. Chart, “Synthesis of model reference adaptive
on the control input can not be known a-priori. Fig. 3 shows systems by liapunov’s second method,” IFAC Proceedings Volumes,
the state trajectories of the plant and reference model. vol. 2, no. 2, pp. 145–152, 1965.

72
Authorized licensed use limited to: University Duisburg Essen. Downloaded on June 12,2023 at 13:45:51 UTC from IEEE Xplore. Restrictions apply.
 [4] S. Sastry, M. Bodson, and J. F. Bartram, “Adaptive control: stability, lti systems with parametric uncertainties,” IEEE Transactions on
convergence, and robustness,” 1990. Automatic Control, vol. 66, no. 11, pp. 5498–5505, 2021.
[5] B. Mirkin and P.-O. Gutman, “Tube model reference adaptive control,” [29] D. Q. Mayne, J. B. Rawlings, C. V. Rao, and P. O. Scokaert,
Automatica, vol. 49, no. 4, pp. 1012–1018, 2013. “Constrained model predictive control: Stability and optimality,” Au-
[6] A. Bemporad, F. Borrelli, M. Morari, et al., “Model predictive control tomatica, vol. 36, no. 6, pp. 789–814, 2000.
based on linear programming˜ the explicit solution,” IEEE transactions [30] A. Dhar and S. Bhasin, “Multi-model indirect adaptive mpc,” in 2020
on automatic control, vol. 47, no. 12, pp. 1974–1985, 2002. 59th IEEE Conference on Decision and Control (CDC), pp. 1460–
[7] D. Garg, M. Patterson, W. W. Hager, A. V. Rao, D. A. Benson, and 1465, IEEE, 2020.
G. T. Huntington, “A unified framework for the numerical solution of [31] R. B. Anderson, J. A. Marshall, and A. L’Afflitto, “Novel model
optimal control problems using pseudospectral methods,” Automatica, reference adaptive control laws for improved transient dynamics and
vol. 46, no. 11, pp. 1843–1851, 2010. guaranteed saturation constraints,” Journal of the Franklin Institute,
[8] D. Q. Mayne and W. Schroeder, “Robust time-optimal control of vol. 358, no. 12, pp. 6281–6308, 2021.
constrained linear systems,” Automatica, vol. 33, no. 12, pp. 2103– [32] J.-J. E. Slotine, W. Li, et al., Applied nonlinear control, vol. 199.
2118, 1997. Prentice hall Englewood Cliffs, NJ, 1991.
[9] F. Blanchini, “Set invariance in control,” Automatica, vol. 35, no. 11,
pp. 1747–1767, 1999.
[10] F. Blanchini and S. Miani, Set-theoretic methods in control, vol. 78.
Springer, 2008.
[11] A. Bemporad, A. Casavola, and E. Mosca, “Nonlinear control of
constrained linear systems via predictive reference management,”
IEEE transactions on Automatic Control, vol. 42, no. 3, pp. 340–349,
1997.
[12] E. G. Gilbert, I. Kolmanovsky, and K. T. Tan, “Discrete-time reference
governors and the nonlinear control of systems with state and control
constraints,” International Journal of robust and nonlinear control,
vol. 5, no. 5, pp. 487–504, 1995.
[13] K. P. Tee, S. S. Ge, and E. H. Tay, “Barrier lyapunov functions for the
control of output-constrained nonlinear systems,” Automatica, vol. 45,
no. 4, pp. 918–927, 2009.
[14] A. D. Ames, J. W. Grizzle, and P. Tabuada, “Control barrier function
based quadratic programs with application to adaptive cruise control,”
in 53rd IEEE Conference on Decision and Control, pp. 6271–6278,
IEEE, 2014.
[15] X. Xu, P. Tabuada, J. W. Grizzle, and A. D. Ames, “Robustness of con-
trol barrier functions for safety critical control,” IFAC-PapersOnLine,
vol. 48, no. 27, pp. 54–61, 2015.
[16] A. D. Ames, J. W. Grizzle, and P. Tabuada, “Control barrier function
based quadratic programs with application to adaptive cruise control,”
in 53rd IEEE Conference on Decision and Control, pp. 6271–6278,
IEEE, 2014.
[17] Y.-J. Liu and S. Tong, “Barrier lyapunov functions-based adaptive
control for a class of nonlinear pure-feedback systems with full state
constraints,” Automatica, vol. 64, pp. 70–75, 2016.
[18] A. L’Afflitto, “Barrier lyapunov functions and constrained model
reference adaptive control,” IEEE Control Systems Letters, vol. 2,
no. 3, pp. 441–446, 2018.
[19] I. Salehi, G. Rotithor, D. Trombetta, and A. P. Dani, “Safe tracking
control of an uncertain euler-lagrange system with full-state constraints
using barrier functions,” in 2020 59th IEEE Conference on Decision
and Control (CDC), pp. 3310–3315, IEEE, 2020.
[20] F. Mazenc and L. Praly, “Adding integrations, saturated controls,
and stabilization for feedforward systems,” IEEE Transactions on
Automatic Control, vol. 41, no. 11, pp. 1559–1578, 1996.
[21] A. H. Glattfelder and W. Schaufelberger, Control systems with input
and output constraints, vol. 1. Springer, 2003.
[22] G. Niu and C. Qu, “Global asymptotic nonlinear pid control with a
new generalized saturation function,” IEEE Access, vol. 8, pp. 210513–
210531, 2020.
[23] J. Alvarez-Ramirez, V. Santibanez, and R. Campa, “Stability of robot
manipulators under saturated pid compensation,” IEEE Transactions
on Control Systems Technology, vol. 16, no. 6, pp. 1333–1341, 2008.
[24] Y. Su, P. C. Müller, and C. Zheng, “Global asymptotic saturated pid
control for robot manipulators,” IEEE Transactions on Control systems
technology, vol. 18, no. 6, pp. 1280–1288, 2009.
[25] S. P. Karason and A. M. Annaswamy, “Adaptive control in the presence
of input constraints,” in 1993 american control conference, pp. 1370–
1374, IEEE, 1993.
[26] E. Lavretsky and N. Hovakimyan, “Stable adaptation in the presence
of input constraints,” Systems & control letters, vol. 56, no. 11-12,
pp. 722–729, 2007.
[27] M. Chen, S. S. Ge, and B. Ren, “Adaptive tracking control of uncertain
mimo nonlinear systems with input constraints,” Automatica, vol. 47,
no. 3, pp. 452–465, 2011.
[28] A. Dhar and S. Bhasin, “Indirect adaptive mpc for discrete-time

73
Authorized licensed use limited to: University Duisburg Essen. Downloaded on June 12,2023 at 13:45:51 UTC from IEEE Xplore. Restrictions apply.