Research documentation · Experimental record
Research

Methodology

Base system

The planned benchmark uses an industrial-flavored multi-agent process-control environment, with the SortingEnv + ContainerGym recycling benchmark identified as the initial candidate.

Fault injection

Faults are introduced through a wrapper around the environment's step() process. The injection layer is intended to be configuration-driven and reproducible.

Recovery layer

Four recovery mechanisms are implemented as swappable modules above the same base MARL system. This preserves the apples-to-apples comparison.

Evaluation

Each mechanism × fault-type combination should be evaluated across multiple random seeds, with confidence intervals where feasible.