The need for more dynamic robots has promoted the development of actuators using high-torque electrical motors combined with lower gear ratios - quasi-direct-drive or proprioceptive actuation. Although providing greatly improved capabilities, torque capacity and density suffer, which is why most applications of quasi-direct-drive actuation have been limited to smaller robots of typically 10-30kg. Using series-elastic actuation allows to alleviate some of these drawbacks. It provides some of the same benefits as quasi-direct-drive while increasing torque capacity and density. However, design and mechanical complexity increases, control bandwidth is limited, and output impedance tends to be higher than for quasi-direct-drive actuators.
Despite the limitations, many small quadrupedal robots with excellent physical capabilities have been introduced using both approaches, e.g. Spot and Anymal. Compared to traditional actuation, this fundamental change in robot actuation design has produced a step-change in their capabilities. Simply put: You cannot control your way out of a bad robot!
And yet, current approaches do not scale well to robots of human-like size and strength, and therefore, their applications are inherently limited. New mechatronic approaches are needed to create robots that are capable of such tasks.
One potential solution comes from nature: The mammalian musculoskeletal system uses complex topologies of (compliant) muscles and tendons, and biarticulation. For example, biological research has shown that biarticulated muscles significantly increase jumping height in humans. In robotics, research has shown that (bi-)articulated elasticity can vastly increase efficiency and peak power capabilities, e.g. for jumping and running.
This assignment focuses on addressing this technological gap in actuation by developing new concepts that use, e.g. arrangements of compliant tendons, biarticulation, redundant actuation, and computational design optimisation. The goal is to develop one such concept (e.g. a leg or biped), ideally supported by an experimental prototype. Depending on the background and interest of the student, the exact direction may be finetuned.