Humans have evolved to run with remarkable adeptness on undulating, natural terrains, a feat that is challenging for even the most sophisticated robots. Uncovering the fundamental principles that underlie our locomotion capabilities has significant impact on our understanding of human evolution and on the design and control of robots, prostheses, and sportswear. However, there are no overground running experiments on undulating terrains that measure how humans run on terrains which resemble those found in the natural world.

We designed an experimental setup using custom-built undulating terrains and conducted human-subject experiments these rough terrains, as well as flat ground. Along with increases in energy expenditure and small changes to stepping kinematics, we find that all subjects very tightly controlled the fore-aft impulse upon landing to reduce scuffing collisions, consistent with the results of our model. Maintaining compliant knee and ankle joints upon ground contact is key to reducing scuffing collisions. We find no evidence for path-planning strategies. This stands in contrast to the heavy use of path-planning strategies utilized by robots for walking and running. Our results are consistent with earlier animal studies of running stability on flat and step-like terrains, and further expands on the prominent role played by intermittent sensing and anticipatory strategies (in the form of controlling scuffing collisions) in maintaining stability.

Most recent conference publication (Adaptive Motion of Animals and Machines, 2019) : How humans run on rough terrain

Imagine playing basketball on an uneven ground. Small variations in the slope of the ground would make dribbling the ball significantly harder. Much like bouncing a ball, running is also the repeated motion of ballistic flight punctuated by a stance phase where the runner redirects their motion into the next flight phase. To understand how running dynamics is affected by terrain slope, we mathematise this caricature of running as a bouncing gait by abstracting the runner to a circular disc in the sagittal plane and specifying simple collision laws of the disc with the ground. Our model expands on canonical simple models of running by incorporating angular dynamics of the runner during flight, since the runner has no control over their angular momentum when they are not in contact with the ground. This can cause the runner to tumble.

Using Monte Carlo methods, dynamical systems theory and probability theory, we find a critical role for reducing tangential collisions with the ground (prevent scuffing collisions at landing) for maintaining body orientation. Foot placement strategies whereby runners land on flat regions of the terrain further aid in reducing instabilities. While no strategy in the absence of feedback control can maintain stability, anticipatory strategies suffice for preventing tumbling falls. Further, we find scaling relations that can be used to assess running stability of animals and robots of varying sizes and morphologies.

For more information, see the published paper: Dynamics and stability of running on rough terrain.