Fuzzy controlled hopping in a biped robot

Current biped robots with articulated legs, even the most impressive to date, still lack the ability to execute dynamic motions such as jumping and running with comparable performance to biological systems. This work explores dynamic jumping with the planar biped prototype KURMET, which employs unidirectional series-elastic actuation at each of its joints. While this actuation scheme enables the performance of high-power dynamic movements like the jump, its presence complicates the jumping control problem and has prevented previous researchers from obtaining precise jump control in systems of reasonable complexity. To manage this problem, this paper develops a layered fuzzy control system for KURMET that realizes repeated dynamic hopping and accurate control of both torso height and velocity at each top of flight. An effective two-stage training approach is used for the fuzzy controller to learn the required, yet highly nonlinear, relationships between its inputs and outputs. Finally, the state machine employed at the lowest-level of control is used to achieve a maximal normalized jump height that outperforms most humans and can be sequenced with the hopping movement.