M.I.L.E.

Human–robot interaction has traditionally focused on settings where humans and robots share a common goal and operate under the assumption of cooperation. While this paradigm has enabled major progress in assistive robotics, collaborative manipulation, and shared autonomy, it often abstracts away from the complexity of real-world interaction.

In natural environments, alignment between humans and robots is frequently incomplete, uncertain, or dynamically changing. Humans may not fully trust, understand, or adapt to robotic behavior, and robots may misinterpret human intent, fail to predict subtle behavioral cues, or operate under conflicting objectives. These forms of partial alignment introduce rich but underexplored challenges for perception, planning, and learning.

The M.I.L.E. workshop (Misaligned Interaction and Learning Engagement Strategies) aims to bring together researchers working at the intersection of robotics, machine learning, cognitive modeling, and human factors to study these challenges in a unified way. We are particularly interested in how misalignment emerges, how it can be detected and modeled online, and how adaptive systems can respond to it in a principled manner.

Topics of interest include, but are not limited to: intention inference under uncertainty, human-aware planning in non-cooperative settings, learning from ambiguous or conflicting feedback, modeling human engagement and disengagement, multi-agent miscoordination, and safety in partially aligned systems.

Beyond technical advances, M.I.L.E. also emphasizes conceptual and interdisciplinary perspectives on interaction, including cognitive science, social psychology, and theories of cooperation and conflict in human–robot teams.

By fostering discussion across these communities, the workshop aims to establish a shared foundation for studying misalignment not as an edge case, but as a fundamental property of real-world human–robot interaction.