Advancing Autonomous VLM Agents via Variational Subgoal-Conditioned Reinforcement Learning | Read Paper on Bytez