NRI: FND: Coordinating and Incorporating Trust in Teams of Humans and Robots with Multi-Robot Reinforcement Learning

点击翻译按钮获取中文摘要

The decreasing cost and increasing sophistication of robot hardware has created new opportunities for applications where teams of robots can be deployed in combination with skilled humans to automate labor-intensive tasks. However, if such systems are to become widely deployable, they must be able to appropriately reason about human teamwork. Therefore, this project will create new methods for generating and solving models for teams of multiple humans and robots working together to solve complex problems. These approaches will be able to learn quickly from limited interactions and consider the dynamic and uncertain nature of coordinating teams of robots and humans. Furthermore, the project will develop methods that allows for communication between the robots and humans and incorporates models of trust to permit humans and robots to appropriately establish trust in each other.In particular, this project will produce several novel methods for modeling and learning solutions for teams of robots interacting with multiple people. The approaches will leverage the strengths of POMDPs to consider the dynamic and uncertain nature of coordinating teams of robots and humans. Because sample efficiency is of utmost importance when dealing with humans and real-world tasks when the number of interactions will be limited, the project will develop Bayesian reinforcement learning methods that scale by exploiting hierarchy and deep learning. The project will also develop methods for communication and shared mental models to allow the humans and robots to have confidence in what each other is doing. These methods will allow tight cooperation between the humans and robots. Furthermore, humans will not want to use our system if they cannot trust the robots. Therefore, the project will develop methods that model and incorporate trust into the approach while generating interpretable POMDP models and solutions that can be shared with humans during or after execution. These advances will produce high-quality solutions for mixed human-robot teams in realistic scenarios.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.