Decentralized self-organizing large-scale systems, such as robot swarms (Hamann, 2018), can be designed to be adaptive, robust, and scalable. However, developing robot and system behaviors that are robust and adaptive to dynamic environments, dynamic system size, and faults is still challenging. We study a swarm showing robust scalability. The swarm needs to detect the change in its environment or in the system size, assess the quantity of the change, and appropriately adapt parameters of its control algorithm. In a second part, we discuss bio-hybrid systems of natural plants interacting with robotic nodes. In the studied example, robotic devices are used to steer the growth of natural plants by exploiting their adaptive behaviors. We use methods of machine learning to model natural plants and to guide their growth and motion with an autonomous system. In biohybrid systems, we can exploit natural adaptive behaviors to build, for example, systems that self-repair.
Swarm robotics is claimed to have the advantage of high
robustness against failures
High degrees of scalabilty can be achieved as swarms rely
on local information and local communication only and
has been shown in experiments
However, if robots break at runtime, then the swarm density changes and the swarm possibly requires online scalability to remain efficient. We require the swarm to be robust against dynamic swarm sizes which can be called ‘robust scalability.’ If the swarm size changes, then each robot may require to change its behavior at runtime, for example, by adapting parameters of its control algorithm.
These measurements are averaged over a limited time window and directly influence control parameters of the robot.
Hence, the robots do not explicitly react to a detected change but forget old measurements that were recorded before the change. Besides measuring light intensities and remembering maximum/minimum light intensities, robots measure the time ta between two encounters of a wall and the time tr between two encounters of a robot. If we assume a regular floor plan (e.g., rectangular) without many obstacles (except for other robots), then the time between encountering walls indicates the side length and hence the area of the room. The time between encountering other robots is similar to a mean free path and indicates the robot density. The distribution of measured times between robot-robot encounters is similar to an exponential distribution and its mean, for example, drops significantly if the robot density is halved. Robots meeting robots on a light spot stay stopped for waiting time w. The time w is a parameter of the control algorithm that depends on these measured values. w is scaled proportionally to the measured time tr. Measured time ta was not used here as the area remained constant in this experiment. The results indicated that robots adapt successfully to a change of system size from N = 10 to N = 5 at runtime and outperform a swarm that was optimized offline for N = 10.
Copyright c 2021 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
Robustness in a Bio-hybrid System
Bio-hybrid systems combine living organisms with
technology. Here, we combine natural plants with robot-like
units. Key of the bio-hybrid approach is that we get
‘lifelike’ features almost for free instead of trying to mimic
behaviors of living organisms in a purely technological
approach. An advantage is, for example, that a bridge built
from living plants reinforces itself and grows stronger over
time instead of decaying slowly but inevitably
In experiment A, we use a simple setup with one plant
(common bean), two lights (left/right), and a camera. We
allow a user to define target points in 2-d space that the plant’s
tip should visit during the experiment
Our tool-chain to solve this engineering task is rather complex. We start from a dataset obtained by preliminary experiments where the two lights are switched on/off in a regular sequence. The plant is photographed every five minutes.
Using computer vision, we extract data that represents the
plant’s reaction to given light conditions and a previous stem
configuration (e.g., length, bend). With that data we train an
LSTM network to obtain a holistic plant model that predicts
the plant’s reaction for a given configuration. We use the
LSTM network and methods of evolutionary computation to
evolve a controller that takes the plant’s configuration, the
light condition, and the user-defined target points as input
and outputs the desired next configuration of the lights. In a
last step, we use the light controllers that performed well in
simulation to control a real plant and find that they succeed
despite an expected reality gap
In experiment B, we support the growth of a group of bean
plants with a scaffold in the form of a diagrid. In the
bifurcation points we place eight robotic devices that use proximity
sensing to detect a close-by plant and that have red and blue
bright LEDs
The task is to grow a user-defined pattern along the diagrid. The pattern is defined and programmed into the robots.
After planting several bean plants, the autonomous system controls the remaining process. A first robot in the userdefined pattern turns on its blue LEDs to attract the plants.
Once detects the first close-by plant, communicates with the next robot of the programmed pattern. turns off its blue LED and turns its blue LED on. This process repeats until the plant approaches the last robot in the sequence and the pattern has been grown. An experiment takes several weeks and was repeated successfully.
In future work, we plan to show a robust bio-hybrid system with the capability of self-repair. We want to scale up one more step (dozens of plants and 18 robotic devices) and to use a more complex scaffold. After having grown the plants for several weeks, we plan to punch a whole into the scaffold. The system is then expected to regrow that part while keeping other areas (e.g., windows) unobstructed.
Hamann, H. (2018).
Springer.
Swarm Robotics: A Formal Approach.