Neurosymbolic Reinforcement Learning with Formally Verified Exploration Greg Anderson UT Austin