On the development and enhancement of artificial intelligence algorithms for swarm robots in real world applications
Abstract
Swarm robotics is an area where using artificial intelligence (AI) can show a great deal of
improvement. Obstacle avoidance, object detection, mapping and navigation are some
the major algorithms required for successful execution of various tasks in the field of
robotics. There is a challenge in applying these algorithms in a manner that swarm
robots can use effectively. These five areas can be further researched to provide a
platform for real world applications. This research aims to tackle the challenges involved
in applying the aforementioned algorithms to swarm robotics and comparing the results
with single robot systems. These techniques can be optimized by leveraging the
advantage of swarm robots communication and scalability. The proposed algorithms
were tested and validated using swarm robots along with profiling and simulations. For
obstacle avoidance, two algorithms were devoloped. The first used a novel and modified
force field method and the second used artificial neural networks (ANN). The results
showed that the modified force field method performed better for static environments
while ANNs worked better for dynamic environments. For object detection, the proposed
algorithm uses an image classifier developed using ANN. The image classifier was
trained to identify blocks of various colours using a convolutional neural network
technique. This algorithm was then applied to swarm robotics using two proposed
methods and results showed that multiple robots viewing objects from different angles
provided better results as compared to single robot systems. This was validated with a
97% accuracy. In two dimension (2D) mapping, the proposed algorithm was developed
using simultaneous localization and mapping (SLAM). The results showed that a single
robot can require upto 3.5x more time for covering a given area compared to a swarm
size of ten robots. This research shows a great deal of contribution in applying swarm
robotics for surveilance purposes by showcasing the ability for swarm robotics to
coordinate and execute the required task in an efficient time frame. The proposed
three-dimension (3D) mapping algorithm used octomaps and occupancy grids to map out
an image taken from a camera mounted on swarm robots. The images were obtained
from various angles using multiple swarm robots. AI algorithms with a focus on swarm
robotics are developed and enhanced for real world applications including fire-fighting,
surveillance, fault analysis and construction. Results showed that swarm robots were
able to complete a given task by up to six times faster as compared to a single robot. The
overall contribution of this research lays a platform for further applications by
showcasing the effectiveness of robotic algorithms in a swarm robot environment.