A Framework for the Design and Evaluation of Robots and AI with Extended Human Adaptation

What is it about?

The problem with collaborative robots (cobots) and AI in the human environment is their lack of ability to adapt to human partners. This research proposes a system that is able to adapt both to immediate changes (like motivation and goals) and long-term trends (like preferences) in people's behavior. Through tests, the system improved the efficiency of cobots and humans working together, making it feel more natural and trustworthy. The authors believe these improvements are important for the future success of cobots and AI, in general.

Featured Image

Photo by Nick Fewings on Unsplash

Photo by Nick Fewings on Unsplash

Why is it important?

First, our framework offers an effective way of adapting to humans, which is a hard problem. The system is able to adapt to short-term human needs and long-term human preferences. Secondly, we outline a full pipeline of developing a human-aware system with training/modeling and experimenting (both in simulation and physical environments). Finally, unlike how AI is evolving at the moment, we believe that using a deep learning model end-to-end for human-aware decision-making is not safe and hard to trust (e.g., black box). So, we add a symbolic layer on top of advanced deep-learning models that are great for *Sensing* the environment. With that, cobots can anticipate and align with human behaviors and adjust their decisions accordingly for a more personalized experience. The robot decisions are transparent (human-like reasoning), interruptable by humans, and incorporate human states/decisions too. That is, using our framework multiple AI model decisions and human decisions can be brought together.