Realistic and Lightweight Cyber Agent Training Environment using Network Emulation in Mininet

In this work, we enhance the CybORG [1] framework to replace its abstract network simulation with lightweight emulation using Mininet [2]. CybORG's abstract simulation provides efficiency for training cyber agents, but its low-fidelity lacks realism needed for training real-world cyber agents. On the other hand, full cloud-based network emulation or live subnets both provide high-fidelity but require additional resources. Our approach is to strike a balance between the two by using the lightweight emulation capabilities of Mininet, thereby simultaneously emphasizing efficiency, cost-effectiveness, and realism. The proposed demonstration highlights the advantages of Mininet-based emulation and its use for cyber-agent training and rapid validation with minimal resource requirements. In addition, our environment enables the collection of network utilization metrics that can further improve cyber-agent training. This work not only broadens the applicability of CybORG for cybersecurity research, but also provides a more realistic, scalable, and cost-efficient environment for cyber agent training.