Strike-vs-Guard: Multi-agent UAV planning and control in adversarial environments

Strike-vs-Guard: Multi-agent UAV planning and control in adversarial environments

Currently open to new students?

Yes
No

Description

Strike vs Guard (SVG) is a multi-agent drone ground control architecture developed at AirLab to explore high-speed coordination and real-time safety in adversarial environments. In this hypothetical "game" setup, two teams, strikers and guards, compete over a protected target, requiring fast tactical adaptation and robust flight control.
We are now expanding to larger team engagements (e.g., 6 vs. 3 games) and developing intelligent multi-agent planning strategies, adversarial game-theoretic controllers, and multi-robot coordination algorithms that bridge simulation to physical flight tests. We are also adding World (Action) Models to the SVG system to learn opponents' dynamics and recognize their intent, informing multi-agent planning and control.
Please also refer to our website:  https://strike-vs-guard.github.io/svg-website/ 

Skills Desired

Required
  • Strong programming skills in Python and ROS2 systems
  • Proactive and fast self-learner
Bonus
  • Familiarity with robotics fundamentals, multi-agent systems, or trajectory planning/control
  • Familiarity with software: Linux, Bash, Git, Docker
  • Familiarity with UAV autopilots, PX4, or motion-capture systems
  • Familiarity with basic mechanical engineering skills (3D printing, SolidWorks), or embedded systems (UART, SPI, IIC, CAN, PCB design, etc...)
  • Interest in motion planning, trajectory optimization, or feedback control
  • Interest in pursuit-evasion games, game theory, control barrier functions (CBFs), or model predictive control (MPC)

Student Learning Objectives

  • Design, simulate, and deploy multi-UAV coordination, pursuit-evasion, and safety-critical control algorithms on physical drones
  • Gain end-to-end systems experience bridging sim, hybrid sim-to-real workflows, and physical flight room testing
  • Learn how to structure safe, distributed autonomy architectures
  • Collaborate closely with senior students and faculty advisors for board aspects of research topics built upon the current multi-drone controller
  • Gain hands-on experience integrating World (Action) Models into a real-time multi-drone system
  • Explore how learned World (Action) Models can predict opponent dynamics and intent to inform planning and control in adversarial games

Classes Accepted into Project

Junior
Senior
Graduate Student
Not a hard limit. We care more about talent and potential.

Compensation

Units
Pay

Contact

(also CC Sebastian Scherer  basti@andrew.cmu.edu )
PS: Due to the nature of this project that requires a wide spread of skills (Mechanical, Electrical, Embedded systems, control theory, planning, perception....), we do not ask you to have all the skills mentioned in the description. Please indicate which aspect most attracts you and where you would like to spend most of your time when you reach out :)