Research Overview

Legged animals outperform legged robots albeit limited muscle power density and sensorimotor conduction velocity responsible for comparably slow reflexes. We see that animals run faster and more efficiently, and incorporate higher agility in a smaller building volume. But how does the animal's neurocontrol orchestrate legs and body, what role plays morphology, and how are animals not overwhelmed by a control task that requires high-end computers, kilo-hertz rate sensory feedback, and complex mathematical models in robots? 

At the Dynamic Locomotion Group, sharing our research results and tools is the goal and we contribute to the open-source robots Oncilla robot [ ] and Solo robot [ ] through the Open Dynamic Robot Initiative. We developed a rugged, light-weight, and customizable foot sensor ('FootTile') to reliably sense touch-down timing and stance phase loading. Arrays of our waterproofed sensor sense the instantaneous center of pressure during locomotion in natural terrain [ ].

We developed a robot leg with animal-like leg segmentation and a modular spring-tendon network. We find that the animal-like spring-tendon configuration hops with an below-average cost of locomotion of animals [ ]. With a hopping robot and boom-guide we recreated an environment that alters the effect of gravity. The gradually changing cost landscape shows how learning of hopping locomotion can be guided through a sequentially increase of environment loading or the robot's complexity [ ].

A close look reveals hysteresis loops in muscles' work cycles also during negative power conditions (active braking or damping). We examine viscous, physical damping that produces forces based on joint velocity without need for high-frequency sensory feedback [ ]. In a numerical simulation study, we provide evidence that small amounts of intrinsic, viscous damping can improve locomotion [ ].

We research the trunk's subtle swaying motion in the fore-aft direction and its consequence to bipedal running. Previously, ground reaction forces were documented that intersect in a virtual point above the human trunk's center of mass. We predict feasible running gaits with a virtual point below the center of mass and characterize related actuator characteristics [ ]. With collaborators from human locomotion research, we then show that human runners display a virtual point below, in certain locomotion conditions [ ]. We applied our model to bird-type and theropod-type trunk morphologies [ ].

Fluid-based (hydraulic, pneumatic) actuation is an omnipresent tool in engineering; less known is that animals and plants integrate fluidic actuation as well. Several spider-leg joints are potentially actuated by hemolymph. Inspired by a highly structured membrane in spider leg joints, we develop a spider-inspired rotary rolling diaphragm actuator [ ] which shows highest mechanical efficiency, independent from actuator speed and load.

With large feedback delays in animals, how can mechanosensory information trigger feedback loops timely? Intra-spinal sensing minimizes feedback delay, and intra-spinal accessory lobes in birds have been hypothesized to act as such sensors. We mapped the topology of the surrounding soft tissues and identified morphologies in small birds (quail) that are candidates to support accessory-lobe-based, intra-spinal sensing [ ]. 

We merge leg designs featuring mechanical elasticities (springs) and active compliance (feedback-and model-based virtual spring) leading to a new class of hybrid joint actuation that allows low control frequencies; the hybrid-actuated leg is controllable with sensory update frequencies around 50Hz, compared to typical, active robot leg compliance at 200Hz and beyond [ ].