Report
ChronoSRL aims to measure time to a goal through embedding distance
Its developer says the self-supervised reinforcement learning method gives its critic a “temporal geometry,” reflecting how a goal close in space can be far away in time.
TLDR
ChronoSRL’s developer says the method is designed so distances between embeddings measure the time needed to reach a goal. The announcement also invites readers to try its policy on a quadruped task and other tasks.
Combined views
3
1 Source, first seen 8h ago
9 reposts
