Abstract:
Animals perform complex locomotion tasks in the presence of large neurocontrol delays and low central command update rate. It is somewhat surprising that legged robots with high control frequencies and microsecond-range control delays are still unable to compete with legged animals in terms of agility, robustness, performance, and energy efficiency.
In the first part of this thesis, we observe the limitations of previous robotic legs and look into the animal leg to draw inspiration and improve robotic leg designs. We observed that when we flex certain joints in the dead animal leg, all other joints move in a coordinated manner, although there is no active central controller. We further observed that birds’ legs structures were loadable or unloadable depending on foot segment orientation. From our observations, we developed a multi-segment leg that behaves like a spring-loaded leg during the stance phase and disengages its elasticity via foot extensor muscle (disengagement tendon) before the swing phase. Unlike previous robotic legs, the robot leg performs locomotion tasks with minimal actuator central control input. Furthermore, this novel robot achieves high speed and efficient locomotion with inexpensive, off-the-shelf hardware based on a Bioinspired leg design. In the second part of this thesis, we take a deeper look into other disengagement tendon tasks under different gait speeds. The disengagement tendon is an extensor muscle for the foot joints. Besides disengaging parallel elasticity, this tendon charges locomotion energy during stance and releases energy during the transition from stance to swing. Therefore, disengagement helps the leg flex faster, have faster leg protraction at high speed, and operate at lower locomotion energy.
In the third part of this thesis, we use our novel leg design into a quadruped robot as a simplified physical model of a quadruped animal. We evaluate the effect of foot joint extensor muscle (disengagement) in quadrupedal trot gait. Unlike bipeds, leg loading is different between the front and hind leg in a quadruped robot. Quadruped animals such as dogs have less mobility in their distal joints for the hind leg compared to the front leg during trot gait. This difference is rooted in the development of extensor muscle at distal joints. We show this difference through our simplified physical model (robot). In walking scenarios, the hind leg foot segment rotates less under the same hip and knee position command due to trunk pitch down during the stance phase. As a result front leg recuperate energy nine times more than the hind leg via their extensor muscle. Since the hind leg does not benefit from the rotation of the foot segment during walking, it can be a source of why the hind leg in animals requires less mobility during walking.
In the last part of this thesis, we investigate the effect of parallel physical elasticity combined with active control in performance robotic leg under central controller delay. Animals locomote robustly and agile despite significant sensorimotor delays of their nervous system. Robot sensors and actuators can respond within a few milliseconds. However, with unknown touch-down timing, controllers of legged robots can become unstable, while animals are seemingly unaffected. We examine this discrepancy and suggest and implement a hybrid system consisting of a physical parallel elasticity and virtual leg length controller. Our system shows robustness in the presence of high sensorimotor delays and low control frequencies.