MAS Multi Agent Systems

ITI's work on MAS technologies focuses on Multi- Agent Reinforcement Learning (MARL): how multiple agents can learn to cooperate or compete with each other to achieve a goal, whether their own or a common goal. We are also applying Imitation Learning and Inverse Reinforcement Learning to MAS. These techniques allow agents to learn complex behaviors from demonstrations, which is essential for handling cases where MARL is difficult to apply directly.

The combination of MAS with Large Language Models ( LLMs ) , where each agent has specific roles (specialization), is another area we are working on. We believe this task specialization can lead to more efficient and accurate solutions in various fields. Furthermore, we are investigating negotiation mechanisms between agents, grounded in game theory and supported by LLMs . This type of project can improve agents' ability to make strategic decisions and cooperate effectively in competitive environments.

Another relevant aspect is the application of Deep Learning technologies , focusing specifically on the use of convolutional neural networks ( CNNs ) in the context of MAS. We believe that CNNs have significant potential to solve a wide range of complex problems. Our goal is to leverage the advanced capabilities of CNNs to develop innovative solutions that can improve the accuracy and efficiency of existing applications, as well as open new opportunities in emerging fields.

In terms of application areas, we're involved in industry- focused projects such as task scheduling , machine cluster management, and the use of IoT sensors . These projects aim to significantly improve operational efficiency and decision-making in industrial environments. Surveillance and land coverage with drones and/or autonomous robots is another area we're exploring, combining mobility and autonomy for real-time monitoring and supervision tasks.

Finally, another potential avenue of work is simulation with digital twins , which integrates MAS and Reinforcement Learning . This approach allows for the creation of virtual replicas of physical systems for testing and optimization, including the use of robotic arms or other automated devices.

In short, we are applying advanced technologies such as Multi- Agent Reinforcement Learning , Imitation​ Learning , MAS- LLMs and Deep Learning in industrial, surveillance, simulation, and negotiation contexts. We believe these projects are not only technically challenging but also have the potential to generate significant impact in various fields of application.