We present the results of an advanced concept research for a collective of robotic rovers named Moonwalkers that implement bio-inspired motivations, distributed intelligence, and integrated deep learning techniques to demonstrate autonomous cooperative goal-directed lunar exploration with minimum human interventions. By borrowing concepts from biology, the Moonwalker rovers are driven by a set of balanced motivations (ex. Energy, safety, operating temperature, achievements, respect) to accomplish mission goals while maintaining safety from potential dangers. Drawing from contemporary neuroscience and computer science research, the Moonwalker rovers are equipped with system 1 (low-level) and system 2 (high-level) cognitive functions based on a two-level memory architecture utilizing a pre-trained multi-modal model in conjunction with a graph knowledgebase, which leverages knowledge graph embeddings. This neuro-symbolic approach enables the rovers to learn from and be guided by human knowledge, facilitating zero-shot learning, collaborative planning, and reasoning like human explorers under uncertain conditions. We also describe the construction of a virtual simulation environment with high-fidelity visualizations and prototype rover agents. By addressing selected problems, we demonstrate that physics informed machine learning can enhance the effectiveness of the virtual lunar test environment.