Distributed autonomous systems testbed

I built and open-sourced a deterministic experimentation framework for autonomous multi-agent coordination. It separates the physical world from each agent’s local beliefs, so a protocol cannot assume information it has not received.

The pipeline keeps communication, protocol decisions, independent safety validation, execution, and post-action reassessment distinct. This makes it possible to examine where a coordination strategy succeeds or fails.

• Centralized and decentralized protocols: compare alternative coordination mechanisms within the same framework.
• Network conditions: configure latency and packet loss to evaluate behavior with delayed or missing information.
• Resource and execution constraints: model limitations that affect which actions are possible.
• Post-action risk checks: reassess the resulting state, including secondary risks introduced by an action.

I developed validated TOML experiment configurations, command-line tooling, parameter sweeps, structured run artifacts, and replayable event traces. Extensible protocol interfaces let coordination strategies change without replacing the entire experiment harness.

Automated tests and GitHub CI support changes to the framework. The combination of controlled inputs and recorded outputs makes it possible to investigate behavior beyond a single successful demonstration.