Traditional robots, such as robotic arms in assembly lines, are designed for specialized, repeatable tasks. However, robotic applications are not always set in structured environments. Rather than using rigid components, soft robots made from soft materials are able to bend, extend, and deform. These movements enable soft robots to swim, walk, crawl, and grasp objects like the impressive capabilities of biological systems, such as the octopus. I hypothesized that the ability of soft legs to easily compress and bend would reduce the complexity of the control algorithms and hardware required for navigating unstructured obstacles. However, because soft walking robots move very differently than their rigid counterparts, they provide a new challenge. Soft appendages must be soft enough to passively adapt to the environment while being stiff enough to generate forces for walking.

In this thesis, I have investigated methods to design, fabricate, model, and control soft-legged robots that are able to navigate over obstacles using simple control strategies. Inspired by nature, I found that a pneumatically actuated soft-legged quadruped robot with three actuated degrees-of-freedom (DoF) per leg was able to navigate over loose rocks or pebbles, squeeze into tight spaces, and walk underwater against flow when augmented with an inflatable soft body. I developed an application-driven design framework to relate the geometry and material properties of the soft legs to common metrics such as bend angle and blocked force. This design framework enables roboticists to rapidly design and fabricate soft robots to satisfy functional requirements. I also developed a lumped-parameter soft robot simulator to replicate the movement of the robot and used genetic algorithms to evolve practical, task-driven control strategies for accomplishing challenging problems, such as squeezing through a confined opening. Practical implementation of these gaits normally requires heavy and bulky pumps and valves which can be very challenging to carry onboard a robot with soft legs. To address this concern, I developed soft air-powered circuits to control the gait of the soft-legged robot without requiring any electronic components. The sequential behaviors were mechanically “programmed” in the circuit by storing information using the snap-through instability in hemispherical elastomeric membranes. The pneumatic memory elements in the circuits changed the walking direction of the robot based on physical interactions with the world. These pneumatic circuits could potentially be used to control electronics-free soft robots for navigating environments where electronics may not be suitable, such as environments sensitive to spark ignition. The contributions in this dissertation enable more versatile soft robotic systems which could potentially be used to monitor hazardous environments.