Extended Reality as a Mediation Layer for Situated Human Control in Human-Robot Teaming

As robots move from isolated industrial settings into dynamic environments like hospitals and collaborative assembly lines, the traditional model of robot autonomy is becoming insufficient. Current human-robot interaction (HRI) often focuses on communicating robot intent—such as where a robot is moving or what it is doing. However, the authors argue that this is not enough. In complex, shared environments, robot plans that are technically feasible may still be socially inappropriate, physically obstructive, or contextually invalid.

To address this, the authors propose conceptualizing Extended Reality (XR) as a 'mediation layer.' This layer facilitates 'situated human control,' defined as the ability of a human collaborator to understand, shape, authorize, and interrupt robot actions in real-time. By grounding robot behavior in the physical workspace, XR allows humans to maintain oversight without needing to constantly monitor low-level robot data.

The authors identify four core functions that an XR mediation layer must perform to support effective teaming:

1. Intent-Autonomy Alignment: Bridging the gap between high-level human instructions (e.g., 'move this') and the robot's specific interpretation of those commands.
2. Plan-Judgment Integration: Allowing humans to evaluate robot trajectories against context-specific criteria like patient comfort or workspace access.
3. Shared Control Management: Making the current level of autonomy—whether the robot is fully autonomous, constrained by a human, or awaiting approval—transparent to the user.
4. Team Roles and Recovery: Managing who has the authority to command the robot and providing mechanisms to recover from plan failures or changing team dynamics.

This research shifts the design focus from merely 'showing' robot state to 'enabling' human-robot collaboration. By treating XR as an active interface for negotiation, developers can create systems that are more accountable and actionable. This is particularly critical in high-stakes environments like bedside nursing or multi-arm teleoperation, where the cost of a misunderstanding between human and robot is high. The proposed design dimensions provide a roadmap for researchers to build systems that allow humans to remain in control, even as tasks and environments evolve.