Abstract

Recent advances in Machine Learning (ML) have led to significant strides in robotics research and interactive robot learning that enable robots to learn from interactions with the environment. However, safety and normative behaviors remain critical factors in deploying these ML approaches in real-world robotic applications in unpredictable, human-populated environments. 

Ensuring physical safety (i.e., avoiding unwanted or harmful contact) has increasingly received attention in control, motion planning, ML, and decision-making research. Typically, robot safety has strictly related to avoiding possible injuries by the robot, but it has become apparent that robots should also be well-perceived to ensure good human-robot interaction (HRI). 

A growing body of work in HRI has shown that cognitive and social safety during interaction can affect both how people perceive and engage with robots, including how susceptible they may be to dark patterns in robot design. However, additional work needs to be done in the robotics community to consider these impacts as well. 

Additionally, robots can increase people's safety by taking over dangerous tasks, like doing heavy lifting or handling toxic spills to prevent negative health consequences for people. However, more research is needed to explore in what contexts and at what times robots can safely take over these tasks, as well as methods to ensure people do not over- or under-rely on them. Safely deploying robots into the real world where they need to work with and alongside people calls for trans- and interdisciplinary efforts.

Workshop Aim

This workshop aims to go beyond the typical notion of robot safety, discussing the challenges of ensuring safety and normative behaviors in real-world HRI, and explore ways to increase safety in and through HRI. We will discuss research methodologies and modeling approaches and propose new and debate existing metrics to accelerate safe and normative HRI research.