By Dr Nidhi Seethapathi.

Even without external perturbations such as pushes or uneven terrain, noise-like imperfections in biological signals continuously perturb animals when they move. In this paper, we inquire how humans run without falling down in the presence of such intrinsic perturbations. We know that such intrinsic perturbations manifest as step-to-step variability and that this variability serves as a metric to quantify stability of movement. However, we understand little about the relationship between this step-to-step variability and motor control. Moreover, human running is most commonly modeled as a spring-and-mass or its variants, which are pertinent for understanding the role of passive actuation in running but don’t explain how deviations from the average motion are corrected. In this work, we explain how humans run without falling down in the presence of intrinsic noise-like perturbations with the help of experiments and investigate the role of active control in stable running in simulation.

We measured step-to-step variability in the motion and ground reaction forces while subjects ran on a treadmill. Next, we mined this variability to understand the relationship between input deviations from the average center of mass states during flight, and output deviations from average foot placement and ground reaction forces in the next stance phase. We discovered that deviations occurring in center of mass velocities at flight are mostly corrected within the next step and this control is tighter for sideways than for fore-aft deviations. Deviations in center of mass height are corrected by shifting the peak of the ground reaction force during stance. The changes in the center of mass motion predict changes in foot placement, in advance of the foot predicting its own placement, and the corresponding time lag allows for the presence of active feedback. We implemented these experimentally discovered control strategies on a simple biped model with muscle-driven actuation and found that it runs without falling down in the presence of large discrete perturbations (see videos below) as well as small continuous perturbations (of the type present in the experiment). Our results suggest that the muscles play an important role in controlling continuous intrinsic perturbations.

The methods used in this work provide a template for analyzing how running is controlled without the need for external perturbations. These methods could be used to further investigate the role of active feedback control by repeating the study with weakened vestibular and visual feedback. These methods could also be applied to different populations, such as athlete vs. non-athlete, to see how they differ in the control strategies used. The real-to-simulation controller we have developed here can be implemented on bipedal robots and exoskeletons to control running using human-derived gains.

Videos: The model with experimentally-derived gains is stable to discrete perturbations (larger than those present in the experiment).

Publication

Seethapathi, Nidhi, and Manoj Srinivasan. “Step-to-step variations in human running reveal how humans run without falling.” eLife 8 (2019): e38371. https://elifesciences.org/articles/38371

About the Author

How do we maintain balance? What is our control strategy during upright standing? These questions are not as easy as they look. Even when standing quietly, we need to activate various muscles in order to not collapse under the influence of gravity. This is made extra difficult by the frequent balance disturbances we experience in daily life, for instance due to neuromuscular noise or people bumping into us. The postural control mechanism must hence be stable, but also adaptive to internal and external disturbances (this adaptability is called postural robustness). Postural control is conventionally thought to continuously actuate the body segments based on joint state; however, this strategy cannot bring forth the bimodal distribution of ankle movements that are commonly observed during standing. Therefore, we hypothesized that balance during quiet standing is controlled intermitted, not continuous. In this study, we demonstrated the relevance of a new concept for postural control called intermittent feedback control, in which each joint is actively actuated only when the instability of the system exceeds a certain threshold.

We used a quadruple inverted pendulum as a model for tiptoe standing to simulate upright posture with internal disturbance accompanied by the change in posture and sensory feedback. The four segments represented the foot, shank, thigh, and head-arm-trunk segments and had human anthropometric characteristics. Each joint was actuated by an anti-gravitational joint torque generated according to three different control strategies: 1) continuous control (a continuous active and passive joint torque based on joint angle and velocity), 2) intermittent control (an intermittent active joint torque only when instability of the system increases and an continuous passive torque), and 3) passive control (only a passive torque from musculotendinous viscoelasticity). Our results show that only when the hip is controlled intermittently, we obtain joint oscillations with amplitudes and variations that are similar to those during actual human standing. Also, there were substantial differences in postural robustness among different joint control strategies.

Figure: a visualization of our model for intermitted postural control

Our results highlight the advantages of intermittency for the postural control system and could have far-reaching implications for training and rehabilitation. For example, based on the concept of intermittent control, we might be able to train the system to be robust to internal and external perturbations of a certain intensity. Furthermore, our findings may provide important insight into a long-lasting debate: is lower variability indeed more stable? In conclusion, our findings suggested that human upright posture could be controlled intermittently and that this intermittent feedback control may be associated with increased postural robustness.

Publication

Tanabe H, Fujii K, Suzuki Y, Kouzaki M (2016). Effect of intermittent feedback control on robustness of human-like postural control system. Scientific Reports 6, 22446. doi: 10.1038/srep22446. http://www.nature.com/articles/srep22446

About the Author

Recent advances in sensor technology allow for the science of gait features to be applied to new services. These services may comprise of e.g. onsite customisation of footwear or garments, sensor-based applications such as activity monitoring systems, and detailed surveillance monitoring. At the Japanese National Institute of Advanced Industrial Science and Technology, we conducted a study to describe sex- and age-differences in gait features of healthy individuals to support others developing services based on gait features.

In this study, we analysed a large dataset of gait in healthy individuals (99 males and 92 females aged 20 to 75) measured in our laboratory. The dataset comprised of 3D positional data obtained using 55 reflective makers and a 3D motion capture system during a 10m overground walk. This dataset is now available online as part of the AIST Gait Database. The AIST Gait Database site is currently only available in Japanese but please contact “dhrc-liaison-ml@aist.go.jp” for assistance in English. We used a principal component analysis (PCA) to identify sex and age effects on walking patterns in the data. PCA can help identify waveform-features from continuous data specific to certain groups, where previous studies disregarded large amount of data and only investigated selected variables at discrete time points. In addition, the waveforms can be reconstructed from the scores of the principal component vectors (PCV), which enabled us to classify a range of different gait patterns. Using this analysis, we identified 6 PCVs which explain more than 5 % of the total variance in the data as shown in Figure 1. Of these, we found a significant interaction between sex and age on PCV 1 and a significant effect of sex on PCV 6, which indicates that these PCVs contain sex differences in the walking patterns. An animation of reconstructed gait with amplified sex differences can be seen in Figure 1 [figure 1(a): PCV1 and figure 1(b): PCV 6].

Our findings advance the understanding of the nature of human gait. We identified clear sex differences in walking patterns, and showed that some of these patterns are affected by ageing while others are not. We believe that these finding are applicable to various health-related services. For example, we can now express gait features of an individual as a score and compare it to a reference group based on the current study’s PCVs. This information might be essential for optimising gait interventions and tracking changes over time. We are now focusing on launching new gait characteristics assessment services for healthy people based on the results of this study.

Figure 1: Reconstructed gait patterns related to (a) PCV1 and (b) PCV 6. (a) Since young females tend to exhibit larger scores on PCV1, PCV1 + 3 SD figures (red line) indicate extremely young female-like gait patterns, and PCV1 – 3 SD figures (green line) indicate extremely male-like gait patterns; (b) Since females tend to exhibit larger scores on PCV6, PCV6 + 3SD figures (red line) indicate extremely female-like gait patterns and PCV6 – 3SD figures (green line) indicate extremely male-like gait patterns.

Publication

Kobayashi Y, Hobara H, Heldoorn TA, Kouchi M, Mochimaru M. (2016). Age-independent and age-dependent sex differences in gait pattern determined by principal component analysis. Gait Posture. 2016 May;46:11-7.

https://www.ncbi.nlm.nih.gov/pubmed/27131170

About the Author

The ability to adapt our walking pattern to the environment is essential for everyday locomotion. This adaptive locomotion is achieved through highly coordinated movements within (intra) and between (inter) our limbs. However, it is not clear what mechanisms exist behind this coordination. Adaptation strategies during walking have previously been examined using split-belt treadmills. In these experiments, the treadmill has two independently controlled belts that force the legs to move at different speeds. In such split-belt treadmill walking, two types of adaptations have been observed: early and late adaptations. Early adaptations appear as rapid changes in inter-limb (e.g. relative phase between the legs) and intra-limb (e.g. stance duration per gait cycle) coordination. By contrast, late adaptations occur gradually after the early adaptations and only involve inter-limb coordination. Furthermore, inter-limb coordination shows after-effects when the belt speeds are equalized. It has been suggested that these adaptations are governed primarily by the spinal cord and cerebellum, but the underlying mechanism remains unclear. To understand the mechanism of these adaptations, we developed a control model based on the physiological findings, and investigated its adaptive behavior via split-belt treadmill walking experiments using both computer simulations and an experimental bipedal robot (Fig. A).

We assumed that the foot contact timing plays a crucial role for these adaptations because previous studies have showed that the vertical ground reaction forces and ankle stiffness remarkably change at foot contact due to changes in the belt speed condition. We developed a two-layered control model composed of spinal and cerebellar models (Fig. B). The spinal model generates rhythmic motor commands using an oscillator network based on a central pattern generator (CPG). It modulates the command timings formulated in immediate response to foot contact, while the cerebellar model modifies motor commands (only affecting the temporal pattern) through learning based on error information related to differences between the predicted and actual foot contact timings of each leg.

Our results showed that the robot exhibited rapid changes in inter-limb and intra-limb coordination that were similar to the early adaptations observed in humans. In addition, despite the lack of direct inter-limb coordination control, gradual changes and after-effects in the inter-limb coordination appeared in a manner that was similar to the late adaptations and after-effects observed in humans (Fig. C). Our results suggest that the modulation of the foot contact timing of each limb could induce the appropriate modulation of the whole body motion (i.e. achieving inter-limb coordination). The model studies are expected to be a useful tool to investigate hypotheses, such as ours, which are difficult to examine from the human measurement experiments.

Publication

Fujiki S, Aoi S, Funato T, Tomita N, Senda K, Tsuchiya K (2015). Adaptation mechanism of interlimb coordination in human split-belt treadmill walking through learning of foot contact timing: a robotics study. J. R. Soc. Interface. 2015 Jul 12:20150542. doi:10.1098/rsif.2015.0542.

http://dx.doi.org/10.1098/rsif.2015.0542

About the Author