AvatARoid: Human-Robot Telepresence in AR
Bridging the embodiment gap between robots and teleoperators through motion-mapped AR overlays
Tech Stack
Overview
Robot-mediated telepresence systems suffer from a fundamental embodiment gap — the remote user's expressions, gestures, and presence are lost when their only proxy is a rigid robot. AvatARoid tackles this by overlaying a motion-mapped AR avatar onto a humanoid robot, giving the teleoperator a richer, more human form of representation. This CHI '25 published research explored whether combining a physical robot with an AR-rendered human appearance could meaningfully improve the telepresence experience for both the local participant and the remote operator.
Process & Approach
The project began with a design exploration phase, surveying existing telepresence literature and identifying the embodiment gap as a core pain point. I developed a Mixed Reality simulation framework in Unity3D that rendered visually and behaviorally realistic humanoid robots in AR, allowing rapid prototyping of different representation modes. Using Azure Kinect for body tracking, I mapped the teleoperator's real-time motion onto both the robot and the AR avatar. A mixed-method user study compared three conditions: robot-only, robot with video feed, and robot with AR avatar overlay. Qualitative interviews and quantitative questionnaires captured social presence, co-presence, and user preference data.
Key Features
- Humanoid robot simulation framework for rapid MR prototyping
- Real-time motion mapping from Azure Kinect to AR avatar
- Three-condition experimental study design (robot, robot+video, robot+AR)
- Mixed-method evaluation combining quantitative scales and qualitative interviews
- Published at CHI '25 — the premier venue in Human-Computer Interaction
Technical Challenges
Achieving convincing real-time alignment between the physical robot's movements and the AR overlay was the central technical challenge. Latency between the Kinect body tracking pipeline and the Unity rendering loop had to be kept below perceptible thresholds. Designing a user study that fairly compared fundamentally different representation modalities — physical robot vs. screen-based video vs. spatial AR — required careful control of confounding variables and a robust counterbalancing protocol.
Impact & Learnings
The research demonstrated that AR avatar overlays significantly enhanced perceived social presence and user comfort compared to robot-only or video-augmented conditions. The simulation framework I built proved valuable for rapid exploration of telepresence concepts without requiring expensive robotic hardware for every iteration. The work was published at CHI 2025, contributing to the growing body of work on mixed reality telepresence.