MAFIS is a fault injection observatory for lifelong multi-agent path finding. Inject faults. Measure degradation. Compare recovery.
scroll
What happens to a live fleet under sustained faults and cascading failures?
Every MAFIS run computes a fault-free baseline instantly, then runs the live simulation with faults active from tick one. The delta between clean and degraded performance is the research output.
When an agent dies, it becomes a permanent obstacle. Agents behind it replan, but their new paths block others, who block others still. MAFIS traces this chain reaction and measures how deep and wide the damage spreads.
MAFIS controls the variables that matter for fault resilience research. Configure a scenario, inject faults, and compare the results against a fault-free baseline, all in the browser.
Built on Bevy. Data-oriented and cache-friendly. Fault injection, heat propagation, and replanning compose as independent ECS systems.
Memory safety at compile time. No garbage collector. Handles 500 concurrent agents in a single browser tab.
Compiles to WebAssembly and runs in the browser. No server, no install. Open a URL, configure a scenario, run it.
Powered by Bevy, an open-source engine driven by its community of contributors.
MAFIS answers one question: how do multi-agent systems behave under faults? Everything else has a home.
For solver comparisons and leaderboards, see the MAPF Tracker.
Visit MAPF Tracker ↗For standardized 2D grid test environments and scenarios, see the MovingAI MAPF Benchmarks.
View MovingAI Benchmarks ↗For bridging the gap between MAPF logic and real-world robotics, see the LSMART testbed.
Explore LSMART ↗
