Many robots are required to move like humans. Human-like motion is useful to efficiently interact with humans and environments built for them, make realistic humanoids or replace actual limbs (prosthetics).

To this end, Artemiadis et al. propose a technique to generate anthropomorphic motion with a robot arm. The task consists in making the robot extremity reach a specific position in 3D in a human-like way. The challenge is that robots with many degrees of freedom can reach a specific point by following many different trajectories (redundant robots).

To choose what trajectory is more human-like, robots observe humans moving their arm (see figure below). This data is then used to create a probabilistic model (Bayesian Network) that describes how joints are related to each other (inter-joint dependencies). These dependencies are taken into account when planning the arm trajectory using inverse kinematics.

Using this technique, robots were able to replicate arm motions previously performed by humans and even generate new ones that had never been observed!