PreprintPDF Available

A Real-World Realization of the AntNet Routing Algorithm with ActivityBots

Authors:
Preprints and early-stage research may not have been peer reviewed yet.

Abstract

To ease teaching self-organizing systems design, we implemented the AntNet routing algorithm for real-world application using educational robots called ActivityBot. Using line sensors and ultrasonic distance sensors, the robotic ants traverse a tiled graph printed on paper, collectively converging to the shortest path. In our descriptions, we address the challenges to face when employing such self-organizing systems on educational hardware and provide a video on YouTube https://youtu.be/JFduHJ0o0UM.
A Real-Word Realization of the AntNet Routing Algorithm with ActivityBots
Jonas Wilfert, Niklas Paprotta, Oliver Kosak, Simon Stieber, Alexander Schiendorferand Wolfgang Reif
,Institute for Software & Systems Engineering—University of Augsburg
{jonas.wilfert; niklas.paprotta}@uni-a.de {kosak; stieber; schiendorfer; reif}@isse.de
Abstract—To ease teaching self-organizing systems design, we
implemented the AntNet routing algorithm for real-world
application using educational robots called ActivityBot. Using
line sensors and ultrasonic distance sensors, the robotic ants
traverse a tiled graph printed on paper, collectively converging
to the shortest path. In our descriptions, we address the
challenges to face when employing such self-organizing systems
on educational hardware and provide a video on YouTube
https://youtu.be/JFduHJ0o0UM.
Index Terms—AntNet, ActivityBots, Teaching
1. Motivation
We present a real-world realization of the AntNet routing
algorithm with small educational robots as agents. AntNet
finds the shortest path on a graph using self-organizing
(SO) ants similar to the ant colony optimization algorithm.
Using such SO algorithms as a new paradigm for devel-
oping software can provide many enhancements compared
to traditional approaches. These can be employed in cases
where the system needs to be robust against failures, e.g.,
in the energy domain [1] or in major catastrophe scenarios
[2]. However, functionality of SO systems are often hard
to grasp. Our real-world realization helps to visualize the
behavior of the individual agents. To ease the entry to this
field of research for future students, we developed a simu-
lation environment to allow testing own implementations of
SO-algorithms. Crossing the gap between a simulation and
reality posed some challenges we address in the following.
While a corresponding video demonstrating our work can
be found on YouTube we also provide our implementation1.
2. Related Work
We implement the AntNet Routing Algorithm [3] which
is a modification of the more general approach of Ant
Colony Optimization (ACO) [4] specialized for finding the
shortest path. Both algorithms are bio-inspired [5] as they
mimic the way that ants in nature release pheromones
while they travel to perform harvesting tasks efficiently.
Similarly, ACO uses several simulated ants to traverse a
graph. Once all forward-ants reach the target node, they
release pheromones relative to the length of the edges they
1. code on GitHub https://github.com/isse-augsburg/AntNetActivityBots
Robotic Ant Routing Table
WiFi Graph-Data GUI
WiFi
Line-Follower
Figure 1. Overview on the system architecture.
took by traversing the route backwards as backward-ants.
At each node, the ant decides which edge to take next
using a heuristic over the pheromone levels and the length
of the next edge. Repeating this procedure several times
reinforces short paths until all ants converge unto a single,
approximately shortest, route. Practical usage of the AntNet
algorithm can be found in the task of routing packages in
communication networks [6]. One major advantage of the
AntNet algorithm compared to classic algorithms like A
is its robustness against failures of nodes and edges in the
network. These can be removed while the algorithm runs and
the SO ants adapt to the situation for finding another short
path. We chose AntNet as an ACO derivative as the concept
of forward- and backward-ants can be applied seamlessly to
robotic hardware.
3. Realization
We split the implementation into two main parts (cf.
Fig. 1). On the hardware side, there are the ActivityBots
(ants) driving on a tiled map printed on paper (cf. Section 3).
Actions of the ants are controlled by the AntNet routing
algorithm running on the central server (cf. Section 3) we
use to emulate the release of pheromones inside a shared
environment. All actors are connected to the same WiFi
access point and communicate over the TCP protocol.
Real World Implementation. The robotic ants are
based on the ActivityBot educational robot produced by Par-
allax [7]. These robots come with various supported sensors,
such as an ultrasonic distance sensor and a line sensor
we used. ActivityBots can be easily equipped with WiFi
and are very maneuverable. This enables them to follow
a graph which we designed consisting of laminated paper
tiles. Inter-connectable tiles represent the edges of the graph.
They show straight and curved lines the ActivityBots can
follow using the line sensors on the bottom of their chassis.
We print two lanes on the tiles to avoid colliding with
oncoming ants comparable to real-world road traffic. The
nodes of the graph are realized as roundabouts to prevent
accidents. Each entrance and exit is equipped with a unique
barcode to assign which node the ant is approaching or
leaving. After reading the barcode using the line sensor, the
ActivityBot checks whether the roundabout is uncrowded
using the ultrasonic distance sensor before entering it. Each
ActivityBot monitors its distance to vehicles in front and
slows down if necessary to avoid collisions. Additionally,
each ActivityBot detects issues like losing the road markings
or entering an unexpected roundabout entrance. If such
failure happens, ActivityBots enter an error state and signal
the misbehavior by blinking with the onboard LEDs until
human intervention. In order to keep the code running on
ActivityBots lightweight and straightforward, all complex
calculations are outsourced to digital twins running on the
central server, we describe in the next Section.
Central Server. A central server is needed to simu-
late the shared environment and synchronize the pheromones
due to the lack of pheromones the robotic ants can leave
behind. The server stores a digital twin of each robotic
ant to allow better control over them. To stay true to the
mindset of the swarm algorithm, the server is not a central
authority with global knowledge but a rather a simulation of
the environment. Each digital twin waits for requests by the
robotic ant when reaching a roundabout and evaluates the
subsequent actions based on the simulated pheromones. The
new instructions are sent back to the robotic ant, which then
executes the requested command while performing some er-
ror checking, as explained in the last section. This approach
has major advantages compared to running computations
directly on the robotic ant: First of all, it allows the ant to
be visualized in a central view running on the server, which
helps to understand the convergence behavior. Furthermore,
it allows the parameters of the AntNet routing algorithm
to be changed or replaced entirely by a different swarm
algorithm without having to reprogram each robotic ant
individually. Due to the physical limitations of the processor
and line sensor we used, the velocity of the robotic ant is
relatively slow. Estimations show that on a graph with seven
nodes and six robotic ants, it would take up to 8 hours to find
the shortest path finally. To decrease this period, the concept
was extended with pure virtual ants. A virtual ant does not
have a robotic ant as a counterpart. Therefore it is not limited
to any maximum velocity, and the number of virtual ants
can be freely chosen. This allows the convergence of the
shortest path to speed up if desired. To synchronize the
virtual with the robotic ants, a change from the original
AntNet implementation was required. In the original imple-
mentation, all ants reach the target eventually. Only then
they also update pheromones on the traversed path with
respect to the respective path length. This approach does
not work with continuous driving ants, as there is a physical
limitation concerning the number of real ants driving within
the target node or on the road without blocking other ants.
As an alternative, we implemented a tick-based concept.
With every tick, the purely virtual ants get moved by a
fixed distance of a few centimeters on the graph, and the
pheromones get evaporated. The evaporation factor needs to
be way longer than usual, e.g., = 0.0005·Nwhere Nis the
number of ants. When an ant reaches the target, it increases
the pheromones on the path relative to its taken path length.
Instead of following the same path back, it swaps the start
and end node and returns to the original start node. The
parameters used in the showcase are α= 1.0and β= 0.1.
Challenges. The line following sensor proved to
be inferior, resulting in high variance of the individual
sensors. Improvements could be achieved by a 3D-printed
light shield and some adjustments to the PI controller.
Additionally, two edge lines have been found to be more
robust than a single following line in the middle of the
road. To ensure the correctness of the barcode reading, parity
bits were added to the barcodes at the roundabouts. Due to
interference of the ultrasonic distance sensors, a randomized
waiting time between the ultrasound impulses was added to
prohibit emergent behavior.
4. Conclusion
We created a visualization of the AntNet Routing Algo-
rithm, which uses real-world agents to illustrate the func-
tionality. Using this algorithm, the agents’ convergence to-
wards the shortest path is very robust against failures, as
seen in the complementary video. Due to the robotic ants’
low velocity, we extended the real-world implementation
with purely simulated ants. Both types of ants seamlessly
work together to optimize a typical graph. By combining
software with hardware, we provide a system that improves
the teaching experience of SO systems by visualizing the
behavior of multiple agents in a real-world environment
while maintaining a reasonable convergence time due to the
simulation environment. This enables students to implement
and test their own SO algorithms in a simple but visually
descriptive way.
References
[1] O. Kosak, G. Anders, F. Siefert, and W. Reif, “An approach to robust
resource allocation in large-scale systems of systems,” in 2015 IEEE
Ninth Inter. Conf. on Self-Adaptive and Self-Organizing Systems. Los
Alamitos, Calif.: IEEE Computer Society, 2015, pp. 1–10.
[2] O. Kosak, F. Bohn, L. Eing, D. Rall, C. Wanninger, A. Hoffmann, and
W. Reif, “Swarm and collective capabilities for multipotent robot en-
sembles,” in Leveraging Applications of Formal Methods, Verification
and Validation: Engineering Principles, T. Margaria and B. Steffen,
Eds. Cham: Springer, 2020, pp. 525–540.
[3] G. Di Caro and M. Dorigo, “Antnet: Distributed stigmergetic control
for communications networks,” Journal of Artificial Intelligence Re-
search, vol. 9, pp. 317–365, 1998.
[4] M. Dorigo, M. Birattari, and T. Stutzle, “Ant colony optimization,
IEEE Comp. Intel. Mag., vol. 1, no. 4, pp. 28–39, 2006.
[5] D. Floreano and C. Mattiussi, Bio-inspired artificial intelligence:
Theories, methods, and technologies, ser. Intelligent robotics and au-
tonomous agents. Cambridge Mass.: MIT Press, 2008.
[6] S. Chandra, U. Shrivastava, R. Vaish, S. Dixit, and M. Rana,
“Improved-antnet: Aco routing algorithm in practice,” in UKSim 2009,
D. Al-Dabass, Ed. Piscataway, NJ: IEEE, 2009, pp. 25–29.
[7] Parallax Inc., “Activitybot 360° robot kit.” [Online]. Available:
https://www.parallax.com/product/activitybot-360-robot-kit/
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
Swarm intelligence is a relatively new approach to problem solving that takes inspiration from the social behaviors of insects and of other animals. In particular, ants have inspired a number of methods and techniques among which the most studied and the most successful is the general purpose optimization technique known as ant colony optimization. Ant colony optimization (ACO) takes inspiration from the foraging behavior of some ant species. These ants deposit pheromone on the ground in order to mark some favorable path that should be followed by other members of the colony. Ant colony optimization exploits a similar mechanism for solving optimization problems. From the early nineties, when the first ant colony optimization algorithm was proposed, ACO attracted the attention of increasing numbers of researchers and many successful applications are now available. Moreover, a substantial corpus of theoretical results is becoming available that provides useful guidelines to researchers and practitioners in further applications of ACO. The goal of this article is to introduce ant colony optimization and to survey its most notable applications
Conference Paper
Full-text available
The objective of this paper is to introduce the Improved-Ant Colony Optimization in Virtual Circuit Networks. We discuss AntNet, an improved version of Ant Colony applied as a routing protocol. A new model is proposed that introduces the application of Ant Colony Optimization in the area of communication network. The proposed model is analyzed with the application of shortest path algorithm in Virtual Private Network. We also discuss the drawbacks of other optimization techniques that paved for the use of Ant Colony Optimization as an efficient optimization technique. In finding the optimal path, the proposed model is found to perform better than the OSPF.
Article
Full-text available
This paper introduces AntNet, a novel approach to the adaptive learning of routing tables in communications networks. AntNet is a distributed, mobile agents based Monte Carlo system that was inspired by recent work on the ant colony metaphor for solving optimization problems. AntNet's agents concurrently explore the network and exchange collected information. The communication among the agents is indirect and asynchronous, mediated by the network itself. This form of communication is typical of social insects and is called stigmergy. We compare our algorithm with six state-of-the-art routing algorithms coming from the telecommunications and machine learning fields. The algorithms' performance is evaluated over a set of realistic testbeds. We run many experiments over real and artificial IP datagram networks with increasing number of nodes and under several paradigmatic spatial and temporal traffic distributions. Results are very encouraging. AntNet showed superior performance under all...
Chapter
Swarm behavior can be very beneficial for real-world robot applications. While analyzing the current state of research, we identified that many studied swarm algorithms foremost aim at modifying the movement vector of the executing robot. In this paper, we demonstrate how we encapsulate this behavior in a general pattern that robots can execute with adjusted parameters for realizing different beneficial swarm algorithms. We integrate the pattern as a virtual swarm capability in our reference architecture for multipotent, reconfigurable multi-robot ensembles and demonstrate its application in proof of concepts. We further illustrate how we can lift the concept of virtual capabilities to also integrate other known approaches for collective system programming as virtual collective capabilities. As an example, we do so by integrating the execution platform for the Protelis aggregate programming language.
Conference Paper
Resource allocation, in terms of balancing supply and demand, is a common problem in many mission-critical systems, such as smart grids. If the participants’ contribution is subject to inertia, they have to solve the problem for a specific time span in advance, that is, create schedules on the basis of demand predictions. To improve the system’s stability, the participants not only have to anticipate the predictions’ inherent uncertainties, but also provide an appropriate amount of degrees of freedom, i.e., reserves, that enable reactive contribution adjustments. Reserves are of particular interest because the problem’s complexity and uncertainty call for rather coarsegrained schedules and an anticipation can turn out to be wrong. In this paper, we present an approach to robust resource allocation in self-organizing hierarchical systems that is flanked by an auction-based algorithm. To deal with uncertain demand, agents learn characteristic prediction errors and create schedules for different possible developments of the demand. This allows the agents to choose the most suitable schedule at runtime. Further, they schedule feedback-driven reserves to be able to cope with unforeseen situations. Based on the example of creating power plant schedules in decentralized autonomous power management systems, we show that our auction-based algorithm clearly outperforms a regio-central approach.
Activitybot 360°robot kit
  • Parallax Inc
Parallax Inc., "Activitybot 360°robot kit." [Online]. Available: https://www.parallax.com/product/activitybot-360-robot-kit/