Embracing Gravity: Rethinking Energy-Driven Modeling, Control, and Design for Soft Robots

May the force of gravity be with you! Efficient locomotion in natural systems emerges not from resisting gravity, but from exploiting it. Human walking, for example, leverages deliberate destabilization of the center-of-mass, converting gravitational potential energy into elastic and kinetic energy through coordinated leg motion. In contrast, many mobile soft robots are designed to maintain stable configurations and tightly controlled trajectories, requiring continuous actuation to counteract their own weight. This increases energy consumption, limits autonomy, and constrains mobility in unstructured environments, posing a barrier to sustained untethered operation in real-world settings. In this dissertation, I investigate how soft robotic systems can harness gravity, elasticity, and inertia to achieve energy-efficient, robust locomotion. Central to this effort is the development of modeling, control, and design principles that enable soft robots to operate outside tightly controlled laboratory conditions. In such environments, uncertainty, environmental variability, and limited sensing are unavoidable. By combining continuum-mechanics-based modeling with control-oriented robotics methods, this work treats soft robots as dynamic systems. Their motion emerges from the interaction between body elasticity, environmental forces, and actuation. The first part of the dissertation addresses modeling. A variational discretization approach is introduced for constructing planar beam models that preserve mechanical structure and energy consistency while allowing different levels of model reduction. These reduced-order models provide a physically meaningful representation of energy flows and support energy-based control design. Building on this foundation, a spatial Cosserat rod finite element framework is developed for soft robots undergoing large deformations. The formulation supports free-floating motion, avoids singular parameterizations, and is compatible with embedded sensing through a strain-based state representation. Combining physical consistency with numerical robustness, the framework enables high-fidelity simulation and model-based control of mobile soft robotic systems in unstructured environments. The second part develops power-efficient control strategies for underactuated and continuum soft robots based on structured energy exchange. Instead of enforcing trajectories that oppose natural dynamics, a two-layer control architecture expresses high-level motion objectives as target energy intervals corresponding to specific dynamic regimes, while a low-level controller regulates total mechanical energy in closed loop. Inspired by swinging control, this approach is demonstrated on an underactuated pendulum-on-a-wheel system, where locomotion emerges from the exchange between gravitational potential and kinetic energy with minimal actuation. The framework is then extended to spatial continuum soft robots modeled as energy-consistent Cosserat rods, where distributed elasticity introduces an additional energy reservoir. Formal analysis establishes convergence of total mechanical energy for smooth dynamics, while numerical studies involving frictional contact and impact demonstrate robust energy regulation and coordinated motion in contact-rich locomotion scenarios. The third part focuses on the embodied realization of the energy-driven framework in soft robotic hardware. A soft-actuated wheel demonstrates gravity-driven locomotion through distributed compliant actuation, where rolling motion emerges from interactions between gravitational, inertial, and elastic energy. To investigate coordination in systems with multiple soft actuators, a modular legged-wheel platform is developed as an open-access experimental testbed for studying actuator orchestration and multi-body locomotion dynamics. Building on these principles, a fully soft ring robot integrates energy-driven propulsion with adaptive morphology in a continuously deformable structure. This design enables dynamic center-of-mass redistribution, large-scale shape adaptation, and traversal of obstacles beyond the geometric limits of rigid wheels. Finally, the actuation architecture is examined through the redesign and characterization of pneumatic segment actuators, demonstrating how efficiency and attainable actuation frequency depend on actuator architecture and influence locomotion performance. To conclude, this dissertation presents an interdisciplinary framework for developing soft robotic systems that achieve controlled motion by working with environmental forces rather than against them. It bridges modeling, control, and hardware design to realize energy-driven movement inspired by natural dynamics. The core contributions include energy-consistent models, control strategies for power-efficient motion, and hardware architectures optimized for adaptive, power-aware operation. Together, these advances enable soft robots to move efficiently and reliably in unstructured environments and lay the foundation for a new generation of systems that harness, rather than resist, the forces around them.