This repository provides an implementation of a resilience-aware Control Barrier Function-based Quadratic Programming (CBF-QP) controller for multi-robot systems. The controller is designed to guarantee sufficient network resilience in a robot networks with distance-based communication models, even in the presence of malicious robots. There exists at most F malicious robots that try to disrupt the consensus protocol by sharing faulty or even adversarial information. Unlike prior approaches that maintain network resilience by either: (1) assuming access to global network state (i.e., knowledge of all robot states), which becomes unreliable when malicious agents are present, or (2) enforcing fixed topologies with known resilience guarantees, our approach enables robots to collectively maintain resilience in dynamic and reconfigurable communication graphs using only locally available information. In our paper, we provide theoretical guarantees showing that - under certain assumptions on adversary behavior - normal robots can achieve consensus despite the presence of malicious agents. The controller also includes additional CBF constraints to ensure inter-agent collision avoidance.
This repository requires numpy, matplotlib, cvxpy, and gurobi. It contains a simulation framework for multi-agent systems with different update scenarios. The simulation allows you to test how agents behave under various attack conditions such as nominal updates, overstatements, and understatements.
simulate.py: Main script to run the simulation.single_integrator.py: Defines agent dynamics and communication models.helper.py: Contains utility functions used in the simulation.
If you run simulate.py, it will prompt you to select a scenario using your keyboard:
Press 1 for Nominal Update
Press 2 for Overstatement
Press 3 for Understatement
To demonstrate our work, we present three different scenarios, whose videos are shown below.
- Nominal Update: Malicious robots share accurate connectivity levels with neighbors.
- Overstatement: Malicious robots overstate connectivity levels with neighbors.
- Understatement: Malicious robots understate connectivity levels with neighbors.
In all scenarios, the robots are tasked with spreading out as much as possible, which inherently conflicts with their need to maintain resilient network formation. Our controller enables robots to spread as far as the desired network resilience is maintained.
As it can be seen from the videos, the type of adversarial behavior directly affects the mobility and performance of the robots. For full theoretical details, implementation insights, and results, please refer to our paper.