A comparative evaluation of human motion planning policies

A recently proposed approach to human intention estimation in goal-oriented motion is further improved in this paper. The main features of the approach are the fact that the unicycle model has been rewritten with the natural coordinate as the independent variable, avoiding the explicit dependence from the walking velocity and lowering the number of input variables to one, the proposal of a novel cost function, weighting not only the energy needed to perform the path but also the current distance of the human with respect to the target, and the adoption of the Frechet metric to assess the similarity of experimental and estimated paths. The proposed cost function is compared with other two cost functions, adopted in previous works, with reference to experimental data. The performance improvements with respect to the previous approach are apparent, either from a qualitative point of view and from a statistical analysis of the error distances.