Collaborative Combat

Our commitment to providing high-quality solutions and excellent customer service has earned us the loyalty of our clients, who continue to rely on us for their reinforcement learning needs. We are grateful for their trust, and we strive to exceed their expectations at every turn.


Collaborative combat represents a fundamental component of military operations, and it continually evolves to meet the ever-increasing demands of our constantly changing environment. In this context, the application of Deep Reinforcement Learning (DRL) proves to be a major asset in optimizing search and interception trajectories. This technology also provides an adaptive response to the specific requirements of defense missions.

In this complex reality, reinforcement learning proves its utility by training agents, such as combat drones, to make real-time decisions. These agents can be trained to analyze situational data, including the positions of enemies, allies, and obstacles, in order to make the most appropriate strategic decisions. Moreover, they can be trained to communicate with other agents to coordinate their actions, share information, and adapt to changing conditions.

In the context of a collaborative combat mission, for example, one drone may be tasked with identifying and tracking a target, while another drone is responsible for providing cover and avoiding obstacles. By using reinforcement learning, drones can learn from their mistakes and adjust their behavior accordingly, resulting in a more efficient and effective approach to combat.

Example of the MMT Project (Man Machine Teaming):

Delfox has excelled in the development of the MMT project, an initiative aimed at creating a deep reinforcement learning system within a complex multi-agent environment. The MMT project led to the development of evasion and interception trajectories, taking into account the constraints of the aerial environment.

The use of an actor-critic approach and suitable neural networks effectively modeled the agents and optimized their performance. Meta-learning based on genetic algorithms added a further dimension of optimization, thereby enhancing the quality of the obtained agents.
Moreover, advanced methods for agent evaluation were developed, ensuring transparency and understanding for all involved parties. These evaluations focus on both agent behavior, describing their strategy, and their internal learning, providing a qualitative description of their capabilities.

Specific project details have not been disclosed to ensure the security of sensitive information and the protection of strategies involved in defense missions.

Flight Demonstration: 

Delfox has demonstrated its expertise by organizing a flight demonstration, highlighting the operability of its agents. The low latencies of our agent models have illustrated the feasibility of this technology in real tactical scenarios involving real drones.

Realmind: Enhancing Collaborative Combat Operations with DRL

The success of the MMT project led to the creation of the Realmind platform, to facilitate the work of machine learning engineers who design and implement autonomous solutions. Designed to model learning agents through Deep Reinforcement Learning (DRL), Realmind offers a strategic response to the challenges presented by the ever-evolving environments characteristic of collaborative combat.

Realmind emerges as an intelligent answer to the defense challenges. Thanks to DRL technology, Realmind actively explores solutions to address complex and dynamic issues. Realmind is a tool designed for engineers to facilitate the implementation, configuration and monitoring of ambitious deep reinforcement learning projects.

To further illustrate this added value, consider a concrete example. In the context of collaborative combat, tactical situations can evolve rapidly, requiring immediate adjustments to minimize risks to both forces and operations. Realmind offers this responsiveness by modeling agents capable of learning, reducing response times and optimizing operational performance.