AppleAnalysisAI ModelsSeptember 17, 2026

Apple's REVERSAL-BENCH measures the reset-free RL cliff

REVERSAL-BENCH controls environmental reversibility via a continuous parameter ρ∈[0,1] and adds a reset oracle across eight manipulation settings in five physics engines. Reset-free agents get absorbed into irrecoverable states as ρ rises, while episodic agents keep learning.

1 source

More stories today

Open the live feed