Project status

Swarm Lab is working. The next job is to connect each robot’s decision code to its own observations and memory, with limits on the messages and computing it can use.

Updated October 4, 2026

Open Swarm Lab →

What works in simulation

• Divide the work. Robots discover tasks, claim them, and handle renewals, expiry, and conflicting claims.
• Pass on discoveries. Direct messages, gossip, and summary forwarding work under message budgets. Temporary coordinators have finite terms and can be replaced.
• Change the setup. Scout, worker, and relay profiles; seeded maps; and fixed or replenishing task arrivals.
• Inspect and repeat. Robot and packet inspection, Python policies, replay, and run exports. Browser and native Python use the same engine; separate local batch jobs can continue after the browser closes.

Movement, radio, and energy are abstract models. A network split can leave robots with competing claims or coordinators.

The work ahead

In progress

Connect the checked decision interfaces to the simulator. Each robot should use its own information and memory, and its choices should stay within its resource limits.

Next · planned

Build a facility-inspection mission with local maps, routes, inspection jobs, and a return-to-hub service cycle. Then run repeatable comparisons against fixed rules, including cases where messages or robots fail.

Later · proposed

Train a compact policy for task and communication choices, explore aerial and mixed ground–air inspection, and build and measure the physical Node and a supported robot adapter.

No learned-policy improvement or physical robot performance has been established. Read the experiment plan →