Soft-bodied animals have relatively few mechanical constraints on movements. This freedom is expected to impose a great challenge to muscle force transmission and body coordination. I have used the caterpillar as a model system to explore the role of soft mechanics in the control of locomotion. Caterpillars are extremely successful herbivores that roam on complex branched structures. They have multiple discrete soft appendages (prolegs) that attach their bodies to the substrate and can be released on demand. This well-defined substrate interaction makes caterpillars ideal for studying force transmission in soft-bodied animals. In this study, a custom two dimensional force sensor array measures ground reaction forces from the caterpillar prolegs during crawling. The data show persistent inter-segmental tensions propagating forward along the caterpillar's body. By loading itself against the substrate, the caterpillar constrains its mechanics to preferentially stretch and achieve locomotion (a strategy I call "environmental skeleton"). While the substrate provides essential support for crawling caterpillars, inching caterpillars have to rely mostly on their hydrostatic skeletons. A field survey of caterpillar gait diversity reveals many different proleg configurations and their associated motor sequences. In caterpillars with partial proleg reduction, we found various intermediate gaits with characteristics of crawling and inching. The transition from crawling to inching seems to require two major evolutionary changes. The reduction of mid-body prolegs allows the body to loop away from the substrate, and strengthening the hydrostatic skeleton prompts the body to flex instead of compress. A model based on tissue properties of Manduca caterpillar suggests that smaller hydrostats are more stable, consistent with the observation that inching caterpillars tend to be smaller. This gait transition was simulated in several soft-bodied robots. From a simple crawling gait, pacing the motor pattern and removing mid body attachment produce an inching gait comparable to inchworm locomotion. Further, my soft-bodied robots demonstrate how nonlinear loading and large deformation result in behaviors that are not sensitive to the variations in motor commands. Clever morphological designs therefore allow us to embed simple adaptive control in the soft structures (e.g. pre-stressing the body for a dynamic event). We expect to find similar control strategies in soft-bodied animals.