Robots operating in extreme outdoor environments face hazardous terrain, unpredictable conditions, and the need for long-term robust operation with minimal human intervention. This thesis proposes a shift from traditional, perception-heavy systems toward robots with embodied intelligence, where mechanical design reduces reliance on complex sensing and control algorithms. Complementary perception and control strategies further enhance their performance.

Three exemplary systems are presented. First, a tethered rover designed for steep agricultural slopes leverages environmental constraints through an anchored tether with an elevated mount for stability. A novel controller prevents slipping and tipping, enabling safe, quasi-static operation and a wide work range. Second, a tethered bicopter designed for safe exploration of glacial moulins exploits dynamic swinging motions to increase its operational range. Its tether provides a power and communication link and allow for stable maneuvering and precise mapping in confined, hazardous environments. Third, GOAT, a morphologically adaptive robot, that leverages environmental computation and compliant interactions during multimodal locomotion in extreme environments. It autonomously switches between driving, rolling, and swimming by reconfiguring its shape, using environmental forces to navigate efficiently with minimal sensing and control. In addition, an extension is presented to enable the robot to reconfigure into a shape suitable for flying.

Together, these systems demonstrate how embodied intelligence can expand robotic capabilities in range, robustness, and adaptability. Validated in both lab settings and field campaigns, including in Japan, Greenland, and the Swiss Alps, these robots show practical potential for real-world applications like environmental monitoring.