Many-Objective Centralized Adaptation Planning: Towards Hybrid Self-Adaptive and Self-Organizing Systems

Full text

Many-Objective Centralized Adaptation Planning:
Towards Hybrid Self-Adaptive and Self-Organizing
Systems
Pia Schweizer
Department of Food Informatics and Computational Science Hub, University of Hohenheim
Stuttgart, Germany
pia.schweizer@uni-hohenheim.de
Abstract—Driving semi-automated vehicles at close distances,
called platooning, emerges as a promising strategy to address
conflicts associated with the ever-increasing traffic volumes on
German highways by optimizing fuel consumption and road
utilization. Mapping the architecture of a self-adaptive sys-
tem to platooning, a fully central coordination of vehicles
would introduce a potential bottleneck, while fully decentralized
decision-making might lead to conflicting adaptations. Therefore,
this project aims to establish a hybrid self-adaptive and self-
organizing system that is robust with micro-level autonomic
adaptation decisions while centrally optimizing the decision-
making.
Index Terms—autonomous systems, optimization, coordina-
tion, adaptation.
I. MOTI VATI ON A ND CHALLENGES
Germany faced a staggering 427,000 hours of traffic con-
gestion in 2023 [1], highlighting not only the substantial
loss of time drivers spent on the road but also the resul-
tant increased fuel consumption, which directly points to
its environmental impact. In the ever-evolving landscape of
automotive technology, platooning, the coordinated driving of
(semi-)automated vehicles in convoys, depicts a promising
approach to address several issues associated with high traf-
fic volumes [2]–[4]. The concept of platooning coordination
can be implemented by applying the architecture of a Self-
Adaptive and Self-Organizing (SASO) system [5]. A SASO
system’s ability of a runtime adaptation as a response to
changes in its environment and the system itself is the outcome
of a coordinated interaction between an adaptation manager
(AM), with its collection of software modules, and its managed
resources (MR), being hardware or software [6]. A general
SASO architecture, as shown in Fig. 1, is constituted of
an adaptation manager AMSASO with a set of goals GSASO.
Depending on the level of control, the AMSASO can be
separated into multiple AMext,l implemented with a structured
functionality, e.g., a MAPE-K model (Monitor-Analyze-Plan-
Execute-Knowledge) [7]. The managed resources MRjare
grouped into subsystems Sj. If the subsystem follows its own
goals Gj, an internal adaptation manager AMint,j takes control
This work is funded by the Deutsche Forschungsgemeinschaft (DFG,
German Research Foundation) – 516601628.
Fig. 1. Architecture for a coordinated SASO system; adapted from [5].
of the respective managed resource MRj. The subsystems can,
in turn, be combined into coordinated resource groups CRi,
where the position of the subsystem and the properties of the
group directly influence the goal achievement. In the case
of fully centralized control of the MRjby the AMext, the
intercalated AMint forwards the respective signals of control.
The focus is on global optimization, with local interests of
the MR being ignored [8]. On the contrary, the AMint will act
as the sole decision maker in a decentralized approach, with
the AMext being non-existent. While this approach considers
local goals, the internal AM is not aware of global optima [9].
By combining both techniques in a hybrid approach, AMext
and AMint collaboratively coordinate the SASO system. By
transferring the concept of the coordinated SASO system to
the platooning example, a single platoon depicts a Coordinated
Resource CRi, composed of numerous subsystems Sj. The
subsystems represent the individual vehicles assigned to ade-
quate platoons by the platooning coordination system (PCS),
constituting the AMext. The PCS aims at optimizing global
goals, e.g., energy efficiency, global safety, road capacity,
and traffic flow [10]. As autonomous vehicles have their
own goals, like improving user comfort and balancing their
cost, they might not be interested in following the commands
of the PCS. Instead, the AMint,j provides them with their
own adaptation commands. The InHOSaS project (’Integrated

Hybrid Optimization of Autonomous Self-adaptive Systems’)
aims to tackle potentially conflicting adaptations by estab-
lishing a hybrid system that combines centralized adaptation
planning, which is robust with local decision-making, and
decentralized decision-making, optimized by central coordina-
tion. For a deeper dive into the approach, Section II outlines
the anticipated research objectives, followed by the evaluation
methodologies in Section III. Finally, this paper closes with
a forecast of the next steps in Section IV.
II. CONTRIBUTION AND OBJECTIVES
While the implementation of a fully central coordination
is computation-intensive, time-consuming, and introduces a
potential single point of failure, the decentralized decision-
making might result in conflicting adaptations with au-
tonomous entities incapable of identifying globally optimal
solutions. Therefore, the InHOSaS research project aims to
develop a hybrid collaborating SASO system. For its realiza-
tion, this project is divided into two branches. The first branch,
addressed by the Intelligent Systems Group at the University
of Kiel, employs a bottom-up approach with local decision-
making based on autonomous learning. The second branch, the
focus of this PhD project, pursues the top-down perspective for
which the following central research question emerges: How
should the central planner be constructed to allow for a
many-objective, self-aware optimization at runtime while
robustly handling the introduction of local decisions that
possibly interfere with its global plan? The development
of a central planner and the subsequent integration of local
preferences can be structured into several steps with individual
research objectives (RO).
RO 1 - Development of a central planner with a
situation-aware, single-objective optimization. With a top-
down perspective, the managed resources receive adaptation
instructions from the central planner and act accordingly.
To achieve coordination optimization at runtime, designing
a modular architecture that involves planning as optimiza-
tion [11] is required. The modularity ensures flexibility and
eliminates the planner’s dependence on a specific optimization
technique. According to the ’No Free Lunch’ theorem [12]
and previous studies on the situation-dependence of various
optimization algorithms [13], no technique performs best for
every objective function or in every scenario. Therefore, a tax-
onomy needs to be empirically created on when to apply which
optimization technique with a focus on single-objectives, con-
sidering stochastic, evolutionary, and mathematical approaches
as such optimization techniques are commonly present in
SASO systems [14]. Furthermore, the highly dynamic na-
ture of traffic requires an automated runtime selection of
the most appropriate optimization technique. Building on the
modularity, the planner will be further supplied with a meta-
adaptation logic that can autonomously select the most suitable
optimization algorithm from a pool of many, depending on the
underlying circumstances.
RO 2 - Expanding the central planner’s focus to multiple
differing goals. A central planner focusing on global optimiza-
tion and considering individual constraints might provoke an
unfair distribution of investments among coordinated subsys-
tems. For instance, in platooning, a platoon leader exhibits a
proportionally higher fuel consumption than a vehicle with an
inner-platoon position. Therefore, it is necessary to establish a
compensation scheme to counteract the unfair distribution of
contributions. Furthermore, the single-objective optimization
is shifted to multi-objective, enabling the focus on multiple
differing goals.
RO 3 - Shift from strict adaptation instructions to rec-
ommendations through degrees of freedom. In parallel with
establishing a situation-aware central planner, the subsystems
of the decentralized setup learn to act autonomously and pur-
sue their interests through reinforcement learning. To ensure
the consideration of both the global optimization as well as the
local decision-making, upon a merge of the two perspectives,
the adaptation plan generated by the AMext will be further
equipped with degrees of freedom, i.e., compiling a set of
recommendations rather than strict adaptation instructions. The
local AMint then chooses its adaptation action by optimizing
the fit to its goals within the pre-set boundaries, which causes
a further shift from multi- to many-objective optimization. In
contrast to multi-objetive optimization, in which the system
tries to optimize a compromise of objectives (e.g., using a
weighted function), in many-objective optimization several
goals are targeted individually.
III. METHODOLOGY
Platooning is the first example of evaluating and comparing
the fully centralized, fully decentralized, and hybrid SASO
system setups. To build on the simulation environment em-
ployed by Lesch et al. [13], the open-source traffic simula-
tion SUMO (Simulation of Urban MObility, [15]) serves to
simulate basic traffic situations. Since SUMO does not pro-
vide a platooning functionality, Plexe [16], with the available
Plexe API for Python, is used as an extension. To cover the
variability of the proposed SASO system, various traffic sce-
narios need to be simulated to reflect traffic’s versatility. The
systems’ performance is evaluated based on different metrics,
such as the energy efficiency and the fuel consumption.
IV. FUTURE WOR K AN D RESEARCH PLA N
The InHOSaS project seeks to create a hybrid SASO system
that combines centralized adaptation planning and decentral-
ized decision-making. This system would address challenges
resulting from the increasing complexity of today’s software
systems through top-down optimization and bottom-up learn-
ing methods. As a distributed multi-agent system, platooning
serves as the initial evaluation domain. Therefore, the first
steps will comprise designing appropriate traffic scenarios
and defining metrics. Subsequently, a classification will be
developed that outlines the suitability of different metrics
depending on the underlying scenario while applying a static
central planner and static local decision-makers, respectively.

REFERENCES
[1] “ADAC Staubilanz 2023: Deutschland-Ticket reduziert Staus nicht,”
Feb. 2024. [Online]. Available: https://www.adac.de/news/staubilanz-
2023/
[2] T. Sturm, C. Krupitzer, M. Segata, and C. Becker, “A taxonomy of
optimization factors for platooning,” IEEE Transactions on Intelligent
Transportation Systems, vol. 22, no. 10, pp. 6097–6114, 2021.
[3] A. Mushtaq, I. u. Haq, W. u. Nabi, A. Khan, and O. Shafiq, “Traffic
flow management of autonomous vehicles using platooning and collision
avoidance strategies,” Electronics, vol. 10, no. 10, 2021.
[4] M. Zabat, N. Stabile, S. Frascaroli, and F. Browand, “Drag forces
experienced by 2, 3 and 4-vehicle platoons at close spacings,” SAE
Transactions, vol. 104, pp. 1173–1181, 1995.
[5] V. Lesch, C. Krupitzer, and S. Tomforde, “Multi-objective optimisation
in hybrid collaborating adaptive systems,” in ARCS Workshop 2019;
32nd International Conference on Architecture of Computing Systems.
VDE, 2019, pp. 1–8.
[6] C. Krupitzer, F. M. Roth, S. VanSyckel, G. Schiele, and C. Becker, “A
survey on engineering approaches for self-adaptive systems,” Pervasive
Mob. Comput., vol. 17, no. PB, pp. 184–206, Feb. 2015. [Online].
Available: http://dx.doi.org/10.1016/j.pmcj.2014.09.009
[7] J. Kephart and D. Chess, “The vision of autonomic computing,” Com-
puter, vol. 36, no. 1, pp. 41–50, 2003.
[8] A. Diaconescu, K. L. Bellman, L. Esterle, H. Giese, S. G¨
otz, P. Lewis,
and A. Zisman, “Architectures for collective self-aware computing
systems,” Self-Aware Computing Systems, pp. 191–235, 2017.
[9] D. Weyns, B. Schmerl, V. Grassi, S. Malek, R. Mirandola, C. Prehofer,
J. Wuttke, J. Andersson, H. Giese, and K. M. G¨
oschka, “On patterns for
decentralized control in self-adaptive systems,” in Software Engineering
for Self-Adaptive Systems II: International Seminar, Dagstuhl Castle,
Germany, October 24-29, 2010 Revised Selected and Invited Papers.
Springer, 2013, pp. 76–107.
[10] C. Krupitzer, M. Segata, M. Breitbach, S. El-Tawab, S. Tomforde, and
C. Becker, “Towards infrastructure-aided self-organized hybrid platoon-
ing,” in 2018 IEEE Global Conference on Internet of Things (GCIoT),
2018, pp. 1–6.
[11] E. M. Fredericks, I. Gerostathopoulos, C. Krupitzer, and T. Vogel, “Plan-
ning as optimization: Dynamically discovering optimal configurations
for runtime situations,” in 2019 IEEE 13th International Conference on
Self-Adaptive and Self-Organizing Systems (SASO), 2019, pp. 1–10.
[12] D. Wolpert and W. Macready, “No free lunch theorems for optimization,”
IEEE Transactions on Evolutionary Computation, vol. 1, no. 1, pp. 67–
82, 1997.
[13] V. Lesch, T. Noack, J. Hefter, S. Kounev, and C. Krupitzer, “Towards
situation-aware meta-optimization of adaptation planning strategies,” in
2021 IEEE International Conference on Autonomic Computing and Self-
Organizing Systems (ACSOS), 2021, pp. 177–187.
[14] E. Henrichs, V. Lesch, M. Straesser, S. Kounev, and C. Krupitzer, “A
literature review on optimization techniques for adaptation planning in
adaptive systems: State of the art and research directions,” Information
and Software Technology, vol. 149, 2022.
[15] P. A. Lopez, M. Behrisch, L. Bieker-Walz, J. Erdmann, Y.-P. Fl¨
otter¨
od,
R. Hilbrich, L. L¨
ucken, J. Rummel, P. Wagner, and E. Wießner, “Micro-
scopic traffic simulation using sumo,” in The 21st IEEE International
Conference on Intelligent Transportation Systems. IEEE, 2018.
[16] M. Segata, R. Lo Cigno, T. Hardes, J. Heinovski, M. Schettler,
B. Bloessl, C. Sommer, and F. Dressler, “Multi-Technology Cooperative
Driving: An Analysis Based on PLEXE,” IEEE Transactions on Mobile
Computing, vol. 22, no. 8, pp. 4792–4806, 2023.