Humans have evolved the ability to walk and move very efficiently. It is hypothesized that when moving humans choose to optimize an objective function or an approximation of it. For example, minimizing the cost of transport (amount of energy to travel distance). In my research I have performed two studies. In the first one I focused on investigating how human choose the walking speed on rough terrain and on level floor relative to their optimal cost of transport (COT). In the second study I focused on a method for simulating and predicting human motion using optimal control.
The thesis is written in the form of two manuscripts:
The first manuscript, Preferred walking speed on rough terrain; is it all about energetics?, accepted to journal of experimental biology and was presented in the 2017 American society of biomechanics conference in Boulder, Colorado. Abstract of the first manuscript:
Humans have evolved the ability to walk very efficiently. Further, humans prefer to walk at speeds that approximately minimize their metabolic energy expenditure per unit distance (i.e. gross cost of transport, COT). This has been found in a variety of population groups and other species. However, these studies were mostly performed on smooth, level ground or on treadmills. We hypothesized that the objective function for walking is more complex than only minimizing the COT. To test this idea, we compared the preferred speeds and the relationships between COT and speed for people walking on both a smooth, level floor and a rough, natural terrain trail. Rough terrain presumably introduces other factors, such as stability, to the objective function. 10 healthy men walked on both a straight, flat, smooth floor and on an outdoor trail strewn with rocks and boulders. In both locations, subjects performed 5-7 trials at different speeds relative to their preferred speed. The COT-speed relationships were similarly U-shaped for both surfaces, but the COT values on rough terrain were approximately 115% greater. On the smooth surface, the preferred speed (1.24+/-0.17 m/sec) was found to be statistically not different (p-value =0.09) then speed that minimized COT (1.34 +/- 0.03 m/sec). On rough terrain, the preferred speed (1.07+/-0.05 m/sec) slower than the COT minimum speed (1.13 +/- 0.07 m/sec) and was statistically significant (pvalue=0.02). Since near the optimum speed the function is very shallow these changes in speed results in small change in COT (0.5%). Thus, this suggests that the objective function when walking on rough terrain includes additional factors to COT such as stability. Some of the finding of the first study led us to think about the idea for the second study: A framework for optimization-based motion prediction of multiple-task manual material-handling jobs. This work was presented in the 2018 International association of ergonomics conference in Florence, Italy. Abstract of the second manuscript:
The ability to predict human motion is a limitation of current digital human model (DHM). One of the main issues when applying DHM is its inability to predict realistic human motion when there is a physical interaction with the environment such as in manual material handling performed as a sequence (denoted multiple-task jobs), or in the case of using an exoskeleton where the device apply force on to the body, thus might change the way that human move. Therefore, conventional method of predication of human motion based on a large data set used to create the prediction based on regression equations has a limitation as to the ability to predicate new interaction. Using previous methods based on optimal control we suggested a framework for motion prediction. The inputs are: the task should be performed, the human anthropometry and a reference experience-based real human motion. Then, based on prior knowledge about ergonomics and bio-mechanics measures that impacts on human motion, using the two optimization steps in our framework, it should output an objective function best describes this task. The two optimization steps are as follows: The first optimization step (the inner optimization) finds the motion that minimizes a specific objective function. Means, this optimization searches the joint trajectories describing the motion that minimizes a given objective function. The second optimization step (the outer optimization) compares the motion generated from the first step to a real human motion and modifies the objective function from the current iteration so that in the next iteration the inner optimization would generate a motion more similar to the reference motion. After this process ends, we obtain an objective function. Assuming that the objective functions of different humans are more or less similar, this objective function might be a prediction equation that enables us to predict the motion of a given task based on the task and worker parameters.
In this study we show a first attempt for creating abilities in this area and merely preliminary results of the inner optimization. We have not reached yet to new scientific findings, but we set some infrastructure for this direction of research. For the development of the inner optimization, the objective function was an equally weighted sum of squared torques and sum of squared jerk, based on the assumption that torque and jerk have a significant effect on the way human moves, and we used one recording of one-hand reaching motion as the reference motion. We tried both numerical and analytical tools. Even though most of similar past works used numerical searches to solve the optimization, we did not manage to find the global optimum using only numerical techniques. We did find the global optimum using analytical techniques. We will present in this work solving the optimization using analytic tools for the sum of squared torques and for the sum of squared jerk separately. The next step would be solving it for a linear combination of the sum of squared torques and the sum of squared jerk.