Learning Adversarial Policies for Swarm Leader Identification Using a Probing Agent

Venue: 2025 IEEE International Conference on Robotics and Automation (ICRA).

Date of Conference: 19-23 May 2025

Abstract:

This study introduces a novel approach to swarm leader identification (SLI) in multi-agent robot systems by employing a physical adversary interacting with the swarm in the same environment. We develop a new simulation environment to study the SLI problem and train an adversary, which we term the prober, to solve the SLI problem using forceful interactions with the swarm as its guiding information source. The prober's policy is modeled using the simplified structure state space sequence (S5) model and trained with the Proximal Policy Optimization (PPO) algorithm. The prober only has access to the information on the relative positions of the other agents. We evaluate our approach through extensive simulations using two performance metrics and validate the sim-to-real transfer through robot experiments. Results on evaluating the performance in 10,000 different testing scenarios demonstrate that our method finds the leader's identity in the vast majority (95.7%) of the cases, regardless of the initial leader selection during training. The proposed system represents the first instance of learning-based automatic identification of leader agents in a swarm. This capability is crucial for enabling efficient and robust human-swarm interaction, understanding artificial swarm behaviors, and analyzing latent behaviors in biological swarms in nature.

Contributions

A PHYSICAL ADVERSARY

Prior work on swarm leader identification assumed a fictitious observer with full visibility. The prober is an embodied agent that shares the environment and must push its way through the swarm to learn anything at all.

PARTIAL OBSERVABILITY

The prober receives only the relative positions of the other agents and its own absolute position. Interaction — counted whenever it comes within a radius R of an agent — is its guiding information source.

SEQUENCE-MODEL POLICY

The policy is built on the Simplified Structured State Space Sequence (S5) model with an embedding layer and separate action and value heads, trained end to end with Proximal Policy Optimization.

SIM-TO-REAL TRANSFER

Deployed as a ROS 2 package on a real robot, with JAX just-in-time compilation running the policy on a Raspberry Pi 4. Belief over the leader collapses to certainty within roughly 13 seconds.

Citation

@inproceedings{bachoumas2025sli,
  title     = {Learning Adversarial Policies for Swarm Leader
               Identification using a Probing Agent},
  author    = {Bachoumas, Stergios E. and Artemiadis, Panagiotis},
  booktitle = {IEEE},
  pages     = {3058--3064}
}

Supported by the National Science Foundation (Grant No. 2014264) and the U.S. Air Force Office of Scientific Research (FA9550-18-1-0221).

Abstract:​

Contributions​