An engineering-design oriented exploration of human excellence in throwing

AbstractHumans are fast throwers, and their bodies differ correspondingly from those of other hominids. One might ask why humans evolved to throw fast while others did not; whether the design of a fast thrower is unique or special and whether indeed humans remain fast within broader comparison sets of non-hominid throwers. As a non-hominid comparison set, we consider a random population of five-link robots with simplified joint angle and torque constraints. We generate 20,000 such robot models and sequentially optimize their throwing motion. Since good initial guesses are needed for each optimization, the robots are first arranged in distance-minimizing sequences in design parameter space. Each robot’s optimal throw then serves as an initial guess for the next one in sequence. Multiple traversals of these sequences, and random perturbations, are used to avoid local optima. Subsequently, regression models are used to predict throwing performance as a function of robot design parameters. From these regression models, the dominant heuristic predictor of fast throwing is found to be a long and light last link. Direct optimization of the robot design leads to much faster throwers, also with long and light last links. In striking contrast, the human arm has two equally long intermediate links of significant mass. Nevertheless, a somewhat human-like arm within the same robot set is found to be a good thrower. On combining several throwing criteria to obtain a single figure of merit, the human-like arm lies in the 96th percentile of the population. Since our human-like arm is a crude approximation of an actual human arm, we suggest that fast throwing by human-like robot arms is not inherently difficult from a mechanical point of view.