Figure 2.

Desired torque profiles defined by the controller for the hips, knees, and ankles. Hip-only assistance was defined by 8 parameters, knee-only assistance was defined by 10 parameters, and ankle-only assistance was defined by 4 parameters, meaning whole-leg assistance was optimized using 22 parameters. For the hips (left) and ankles (right), torque (black) was commanded as a function of time, defined as a spline fit to nodes (red) that were optimized during the experiment. For the knees (center), torque was commanded both as a function of time (black), joint angle, and joint velocity. During stance, the knee torque was a function of knee angle to mimic a spring (red), where the spring’s stiffness was optimized. During late swing, torque was a function of knee joint velocity to mimic a damper (red). The red curves shown for these periods of state-based control are the average applied torque at the knees from whole-leg assistance, but the applied torque could vary based on the user’s kinematics. The steep increases and decreases in applied torque during knee extension assistance were due to the impedance controller being turned on and off as a function of percent stride defined by the nodes while the knee angle was nonzero, resulting in discrete jumps in desired torque.