While modern artificial intelligence agents have achieved superhuman performance in specific tasks, training artificial agents that can efficiently explore and generalize to new tasks remains an open problem. Recent work has turned to humans as a source of inspiration to tackle this problem. Humans have shown an ability to explore the world in such a way that translates to generalizable skills in later life, a trait that modern artificial intelligence has failed to replicate. Investigating the specific ways in which humans and artificial agents deviate in behavior may thus lend insight into ways that algorithms can be improved. In this work we present an online platform to design and carry out experiments comparing human and agent behavior in 3D navigation tasks. We also present a comparison of behaviors between humans and agents in both procedurally-designed and human-designed mazes, highlighting ways in which current algorithms are both similar and distinct from humans.