We are developing a control architecture for bipedal locomotion devices such as robots and powered orthotics. Advanced nonlinear control techniques, including feedback linearization, sliding control, and multivariable optimization, are utilized in this control architecture, yielding a highly stable and tunable controller for a highly unstable and nonlinear plant. Tests�using a 3-D, 12-degree-of-freedom humanoid model�include a variety of disturbed initial states and of control goals for the center of mass, swing foot, and other points being controlled. An interesting property of this controller is the emergence of appropriate non-contact limb behavior in response to disturbances. Also, due to its large range of operation, this control architecture can reject significant disturbances more easily than simpler controllers, and requires a less-detailed reference trajectory than simpler controllers. This has the additional benefit of reducing the computational workload of a motion planner in an integrated motion planning and control system. Such control architectures will find use in assistive devices for the elderly and handicapped.