Failure Mode Analysis

Risks & contraindications

Evidence quality

Highlighted claims

• In dense plumes, agent failures cluster near the odor source, indicating overshooting as the dominant failure mechanism. — Clock-state olfactory search in turbulent flows using Q-learning: The geometry of plume recovery
• As plume sparsity increases, failures shift progressively to locations far downwind of the source, indicating agents become trapped in informationally sparse regions. — Clock-state olfactory search in turbulent flows using Q-learning: The geometry of plume recovery
• A single-Q agent trained in one sparsity regime lacks the behavioral repertoire to counter the dominant failure mode of a different regime, resulting in poor cross-environment generalization. — Clock-state olfactory search in turbulent flows using Q-learning: The geometry of plume recovery
• Ablation of the two-Q architecture confirms role separation: using Q-minus alone leads to frequent failure in the back of sparse plumes, while using Q-plus alone increases failure near the source. — Clock-state olfactory search in turbulent flows using Q-learning: The geometry of plume recovery