Research documentation
·
Experimental record
Research
Base environment
The research proposal identifies an industrial-flavored MARL benchmark as the intended base environment. The initial candidate is the SortingEnv + ContainerGym recycling benchmark.
Role in the benchmark
The environment remains constant while the fault injector and recovery mechanism are varied. This isolates recovery behavior as the main experimental factor.
Implementation boundary
The fault-injection layer should sit around the environment's step interface so observations, actions, communication, agent participation, and policy behavior can be altered in controlled experiments.