Creating autonomy for social robots

Description

The technologies enabling social robots are becoming more capable and affordable. Yet creating autonomy for human-robot interaction is still a nontrivial task, requiring the integration of essential AI building blocks, including natural language, vision, navigation, and planning and reasoning. In particular, it’s often difficult to quantify human social behavior and rules with algorithms in robots.

Research has centered around deploying social robots to interact with real people in public spaces such as train stations and shopping malls. Building on over 20 years of combined experience with robots ranging from robotic wheelchairs to highly realistic humanlike androids, Dylan Glas and Phoebe Liu explore some of their work using AI and machine learning to understand social situations and generate robot behavior and using learning by imitation techniques to reproduce social behavior in an autonomous robot, including ways they have adapted robots to the challenges of real-world environments outside the lab, such as using sensor data to understand the social use of space and predict the movement of pedestrians. You’ll see the algorithms and architectures Dylan and Phoebe have developed to enable social robots to approach people, hold conversational interactions, and provide guidance and services to people in shopping malls, elementary schools, and beyond.

They also detail a data-driven approach for generating speech and motion behavior in an autonomous robot, using learning-by-imitation from unscripted human-human interaction data collected from sensors. This technique requires no manual design, human annotation, modeling, or natural-language understanding, and it’s able to autonomously generate socially appropriate locomotion and speech behaviors for a humanoid robot. You’ll leave with some key insights into how learning from human behavior is critical for developing autonomous robots for the real world.