Natural gait discovery for compliantly actuated legged robots

Kurzfassung

Nature evolved legs to travel in difficult terrain. Biological legged locomotion displays different gaits to minimize the energy needed to move with certain velocities. For example, humans walk with low velocity but switch to running when moving faster. Robotic legged locomotion proves to be a challenging topic. Most state-of-the-art motion planning approaches heavily control legged robotic systems in order to perform aspired movements. Changing the intrinsic dynamics of the system through control is necessary since the dynamics do not inherently exhibit the desired motions. However, recent work suggests that embedding desired dynamics into segmented robotic legs reduces the necessity of extensive control. This Master's Thesis combines theoretical templates for legged locomotion with physically realizable robotic legs. It first presents energy-conservative spring-loaded inverted pendulum (SLIP) models, which explain the existence of different gaits in legged locomotion. These templates are modified to capture the effects of physical damping and ground contact dynamics. A minimal control action compensates the resulting energy losses. Hereby, several natural gaits could be displayed with a non-conservative SLIP model. Based on this newly developed model, a segmented leg featuring a pantograph mechanism is designed. The inherent dynamics of this robotic leg are matched with the SLIP dynamics of the non-conservative template model. Based on optimized design parameters, the influence of deviating parameters on the natural dynamics is analyzed. These imprecisions occur when physically realizing the robotic leg. Natural gaits can be transferred from energy-conservative SLIP models to non-conservative models. SLIP-like dynamics are embedded into segmented legs, which can reduce the control necessary to exhibit periodic motions. Gaits supported by the intrinsic dynamics result in reduced energy consumption.