A central topic of my PhD research is to introduce a novel method for representing, generalising, and comparing gaits; particularly, walking gait. Human walking gaits are a result of complex, interdependent factors that include variations resulting from embodiments, environment and tasks, making techniques that use average template frameworks sub-optimal for systematic analysis or corrective interventions. The proposed work aims to devise methodologies for being able to represent gaits and gait transitions such that optimal policies that eliminate the inter-personal variations from tasks and embodiment may be recovered.

Our approach is built upon work in the domain of

  • learning by imitation
  • null-space policy learning
  • operational-space control
  • gait analysis

Representation of human gait

We assume that the human gait we observe is the combination of the consistent characteristics of the gait and variations from environment, embodiments, and behaviours. For instance, people normally walk with difference speeds and cadence, but there must be some underlying consistency so we can recognise walking behaviours as belong to the same class of human gait.

Based on the formulation in operational-space control, we proposed a method such that the observations can be decomposed into the consistent characteristics and variations.


Generalising movement across various behaviours

What defines a human movement is an overloaded question with contradictory interpretations. One way to arrive at a solution is to employ optimisation strategies that somehow capture the essence of gait. However, this is not trivial since the motion that we observe are masked by intra-personal and inter-personal variations.

For instance, walking can be affected by speeds (intra-personal) and body size (inter-personal). The following figures show the observed trajectories of hip joint-angle using motion capture. Although the subjects were asked to walk in a normal way, there are natural variations that arise from different behaviours.

Hip angle of five different subjects walking at one meter-per-second Hip angle of a subject walking with five different speeds

We propose a novel method for generalising the consistent characteristics of human movement subject to variations in environment, embodiments, and behaviours. This is achieved by reconstructing an unconstrained policy without explicit knowledge of the variations.


Adaptation to Unseen Constraints

Many everyday human skills can be considered in terms of performing some task subject to a set of self-imposed or environmental constraints. As an example of a wiping robot, the behaviour (wiping) is subject to various constraint imposed by the environment where the behaviour is performed (surfaces).

Our aim is to develop a method such that some previously learnt behaviours can be adapted to new environment in an appropriate way. In particular, we consider learning the null space projection matrix of a kinematically constrained system, and see how previously learnt policies can be adapted to novel constraints.


Quantify the difference between gaits

One of our objectives is to measure the difference between gaits. A potential application in gait rehabilitation, for instance, is to determine how much the device should correct the subject. The principle is to compare the subject’s gait with an appropriate reference gait which is expected to be normal and use their difference in a feedback controller. For this, we need a way to compare the difference between two gaits.

However, it is sub-optimal to compare the distance between two sets of observations directly, since the observations contain natural variations. A more appropriate approach is to quantify the distance between the characteristics of two walking gaits and ignore the differences coming from these variations.

We propose a procedure to quantify the distance between gaits; at meanwhile, ignores the differences coming from variations. Specifically, this is done by comparing the constrained policies between a walking gait and a reference gait without explicitly knowing the variations. (e.g., the walking gait in question and the reference gait can be a mobility-impaired patient and the healthy subjects, respectively.)

Validation with motion capture data

We explore the utility of our approach on human walking data. Our analysis is based on kinematic and kinetic features of subjects walking with various speeds. Our goal is to see whether we can
  1. extract consistency across walking across different behaviours and subjects
  2. use the learnt gait to quantify the difference between normal and pathological gaits
A snapshot of experiment in progress. A subject is wearing Xsens motion capture system and walking on the V-gait system.

Last update: 08/10/15