We are motivated to better understand how adaptive locomotion (rough terrain locomotion, turning, gait transition, etc) can be realized using a quadrupedal platform with constrained resources. These constraints include computational power limitation, no accurate force/torque sensing, and partial sensing of robot’s kinematic states. These constraints arise from the fact that we are designing and experimenting with autonomous light-weight and (comparatively) cheap quadruped robots. The practical benefit of such robots is fast experimentation: experiments can be safely done with presence of one or two humans, and repairs are cheap and quick.