What this paper is about
This ICRA 2023 workshop paper compares three different approaches to learning heterogeneous robot swarm operations — swarms made up of robots with different capabilities, rather than identical units — in abstracted, complex adversarial environments. It builds directly on the swarm-tactics learning framework explored in Learning Robot Swarm Tactics over Complex Adversarial Environments, extending the question from “how do we learn tactics for a swarm” to “how does the best learning approach change when the swarm isn’t homogeneous.”
Heterogeneity matters in practice because most real swarm deployments — a mix of aerial and ground robots, or robots with different sensors and payloads — aren’t uniform. A learning approach that works well for a swarm of identical robots doesn’t automatically transfer to a mixed team, since the policy now has to account for which kind of robot is making each decision, not just how many robots there are.
Full write-up in progress — check back soon, or get in touch if you'd like the paper directly.
- What is a heterogeneous robot swarm?
- A heterogeneous swarm is a team of robots with differing capabilities — for example, a mix of aerial and ground vehicles, or robots with different sensors — as opposed to a homogeneous swarm of identical units.
