Title: “Efficient height map learning for traversability estimation and footstep planning.”
To this day, humanoid robot locomotion most commonly follows a static walking approach that ensures stability of the robot at any time but also greatly limits the step distance. If any attempt to a more human-like dynamic walking strategy should be made, it is crucial to have a fast and reliable method for extracting necessary information about the local environment in order to be able to plan safe motions. In this talk, we present our approach that uses RGDB sensor data from a camera mounted on top of a humanoid to create a local height map at a high frequency. We analyze the map for traversability and constantly update the robot’s movements if necessary. Unlike other state-of-the-art approaches, our framework is capable of directly reacting to moving objects or sudden changes of the goal direction by recalculating footsteps in a matter of milliseconds.
Maren Bennewitz is a professor for Computer Science at the University of Bonn and head of the humanoid robots laboratory. Her research focuses on robots navigating in human environments. She has contributed probabilistic methods for 3D environment modelling as well as for manipulation and navigation with wheeled and humanoid robots.