The Women in AI speaker series, a collaboration between the Schwartz Reisman Institute for Technology and Society and Deloitte, welcomes Maren Bennewitz, a professor for humanoid robots at the University of Bonn whose research focuses on robots acting in human environments. Bennewitz has developed innovative solutions for robotic systems co-existing with humans, including machine learning techniques for navigation, detection, and prediction.

In this talk, Bennewitz will explore her solutions work utilizing convolutional neural networks and reinforcement learning to enable service robots to act with foresight when navigating human environments.

Abstract:

Service robots acting in human environments need to be able to navigate in challenging scenes, act foresightedly, and avoid interferences with the users while respecting their navigation preferences. In this talk, I will first introduce our framework to enable humanoid robots to navigate through cluttered scenes. Our method exploits knowledge about different obstacle classes and selects appropriate robot actions. To classify objects from RGB images and decide whether an obstacle can be overcome by the robot with a corresponding action, e.g., by pushing or carrying it aside or stepping over or onto it, we train a convolutional neural network (CNN). As I will demonstrate in our experiments, using the CNN the robot can robustly classify the observed obstacles into the different classes and exploit this information to efficiently compute solution paths. In the second part of my talk, I will present an approach to predict human navigation goals based on learned object interactions. I will then show how this knowledge can be used by a robot to realize foresighted navigation in service robotic applications. Finally, I will introduce a reinforcement learning framework to train a personalized navigation controller with an intuitive virtual reality demonstration interface. Our user study provides evidence that the personalized approach significantly outperforms classical navigation approaches with more comfortable human-robot experiences.