ADHERENT: Learning Human-like Trajectory Generators for Whole-body Control of Humanoid Robots

Human-like trajectory generation and footstep planning represent challenging problems in humanoid robotics. Recently, research computer graphics investigated machine-learning methods for character animation based on training human-like models directly motion capture data. Such proved effective virtual environments, mainly focusing visualization. This letter presents ADHERENT, a system architecture integrating used with whole-body control employed robotics to generate stabilize xmlns:xlink="http://www.w3.org/1999/xlink">human-like trajectories robots. Leveraging human locomotion data, ADHERENT yields general planner, including forward, sideways, backward walking that blend smoothly from one another. Furthermore, at the joint configuration level, computes data-driven postural reference coherent generated footsteps, thus increasing xmlns:xlink="http://www.w3.org/1999/xlink">human likeness of resulting robot motion. Extensive validations proposed are presented both simulations real experiments iCub robot, demonstrating be robust varying step sizes speeds.