In this paper, we present an energy-efficient control strategy for legged robots for exploration in low-gravity environments (e.g., the Moon, Mars, and asteroids). Considering limited energy and heat budgets, we propose a reinforcement learning-based control scheme that uses an energy-optimized reward function tuned to gravity magnitude. We develop and validate walking and basic attitude controllers in various gravity environments, from lunar gravity (1.62 m/s²) to simulated hyper-Earth gravity (19.62 m/s²). In Earth gravity, a 15.65-kg robot achieves a power consumption of 23.4 W at 0.4 m/s, a 23% improvement over the baseline policy. In lunar gravity experiments, we achieve a power consumption of 12.2 W, a 36% reduction compared to the baseline controller. This study provides a scalable approach for developing energy-efficient walking controllers for legged robots in various gravity levels. Furthermore, we design a force-compensation spring system for conducting practical experiments in lunar gravity.
