ChronoSRL Introduces Temporal Geometry to Self-Supervised Reinforcement Learning

ChronoSRL equips critic embeddings with explicit temporal geometry by training embedding distances to match the actual time taken to reach a goal. This method incorporates goal-reaching time and survival reinforcement learning to better predict the reliability and distribution of goal attainment.