Researches

Mixed Reality based Multi-Agent Robotics Framework for Artificial Swarm Intelligence Experiments

Technological Advances in Science, Medicine and Engineering Conference (TASME) - Abstract Publication

Swarm Intelligence is the collective behaviour of many individuals toward a certain task, and each operates autonomously and in a decentralised manner. As the world is heading towards a future where automated robotic agents will be carrying out complex tasks without human intervention, research on improving the collective intelligent behaviours of robots has gained significant importance. Experimenting with a swarm of physical robots is challenging since swarm intelligence experiments demand a large number of robotic agents, typically prohibitively expensive, requiring special hardware and a well-controlled environment.

As an alternative, computer-based simulations with virtual agents have been predominantly used in the literature. Simulations are easier to set up, less expensive, typically faster, and more convenient to use than physical swarms. The two main advantages of virtual robot simulators are the flexibility to add new features and efficiency. Virtual simulator frameworks for multi-agent systems may integrate individual behaviour models, communication and interaction models, and programming models. However, the persistent issue with the pure virtual approach is that the simulators cannot always represent the intricacies in real-world experiments, how they react to complex physics, noisy sensor data, control loop delays, etc.

This research aims to introduce a mixed reality framework by merging the two realities and modelling swarm behaviours in such a way that both real and virtual robotic agents, as well as other environmental elements, can co-exist and interact with each other. In this approach, all real robotic agents and environmental elements in the system are simultaneously represented in virtual reality, while the simulated environment may contain various additional elements that are purely virtual. Thus, this paves the way for the creation of safer and lower-risk environments for extensive testing of swarm robotics behaviours, as well as an unrestricted means of achieving agent and environment expansions that are physically impossible or expensive in reality.

As the outcome of this study, a modular and customizable framework is introduced, with the ability to handle the inter-reality sensing and communication and run the swarm behavioural experiments with both real and virtual robots at the same time in the same environment, with different physical and virtually defined environments.

Gallery

Authors

Dilshani Karunarathna
Nuwan Jaliyagoda
Dr. Isuru Nawinne
Prof. Roshan Ragel

Resources/ Links


Next