Walking on New Ground - Toward Legged Robots for Planetary Exploration

Robotic planetary exploration is critical in expanding our understanding of the solar system and preparing for future human missions to the Moon, Mars, and beyond. Traditionally, such missions have relied on wheeled rovers, which are proven to be robust but limited in their ability to traverse complex terrains such as steep slopes, loose regolith, and rocky surfaces in low-gravity conditions. Moreover, these systems typically require separate robotic arms for manipulation tasks, which increases their mechanical complexity and mass. Legged robots offer a promising alternative: they can adapt to diverse terrain and gravity levels and use the same limbs for both locomotion and manipulation. This thesis advances the capabilities of legged robot prototypes for planetary exploration by addressing three key areas: mission integration, low-gravity locomotion, and manipulation. First, we investigate the integration of legged robots into planetary analog missions, focusing on system integration, communications, and operational workflows. We deployed legged robots in multiple analog environments and evaluated their performance in representative exploration tasks. To this end, we evaluated the scientific data return of our robots in analog missions. Additionally, we conducted a comparative study between single-robot and multi-robot deployments, highlighting the operational advantages and limitations of each setup in realistic mission scenarios. Second, we address low-gravity locomotion with reinforcement learning-based control policies. We designed controllers for jumping, in-flight attitude control, and landing, to enable legged jumping locomotion in very low-gravity environments. We successfully validated this approach in microgravity during a parabolic flight campaign. Furthermore, we developed an approach for efficient legged locomotion in multiple gravity levels. To this end, we introduce a gravity-scaled power reward structure that allows training efficient policies for different gravity levels. Third, we develop manipulation capabilities using pedipulation, a technique that uses the robot’s foot for interaction tasks. This approach enables basic manipulation without dedicated arms, which reduces system complexity and mass. We validated our pedipulation controller during a planetary analog deployment on Mt. Etna, where the robot performed terrain probing and sample collection in unstructured volcanic terrain. Additionally, we demonstrated pedipulation in Earth-relevant tasks such as opening doors, pushing buttons, and moving a backpack, showcasing its versatility across domains. Together, these contributions represent a significant step toward deploying legged robots in planetary missions. This work lays the foundation for their integration into exploration missions by demonstrating their versatility and effectiveness in analog environments.