Bio-inspired controllers for dynamic walking: CPG enhanced with minimal feedback

Many of today’s successful biped robots are controlled based on the Zero Moment Point (ZMP) method, which generates robust locomotion for flat-footed bipeds on a controlled environment. These controllers usually generate slow gaits with a low energy efficiency which do not look natural or human-like. In contrast, Passive Dynamic Walkers [2], [5] are fast, efficient and resemble human walking. Motivated by these capabilities, different passivity-mimicking dynamic walkers [1], [2], [3], [8] have been developed. Most published works deal only with dynamic walking over flat terrain and only a handful investigate the effects of varying slopes on gait stability, usually remaining within the range of ±5◦ even when using complex models with knees or a torso. Goswami et al. developed a passivity mimicking controller that uses continuous feedback to maintain a predefined energy level and progression speed. This controller allowed a compass biped model to walk on slopes of up to +12◦. Stable, dynamical gaits that are robust to disturbances can be generated by central pattern generators (CPGs), as shown in studies of animal locomotion [6]. CPGs are networks of coupled oscillators whose outputs are converted into control signals that drive the different muscles. CPGs are especially suited to control legged robots since they can take advantage of the robot’s inertia and allow for free dynamic motion of the robot, utilizing minimal periodic actuation during the gait.