Anyone who has been taught a physical skill knows the feedback problem. The teacher says your elbow dropped. You didn’t feel your elbow drop. By the time you watch the video, you’re somewhere else, watching a flat rectangle of a body that doesn’t feel like yours, trying to map a correction back onto a sensation you can’t recall.
FenceXR (arXiv, this week) attacks that gap with augmented reality movement replay: an error-detection training system that provides spatially grounded feedback by replaying the learner’s own movement in the space where it happened.
Why “spatially grounded” is the operative phrase
The replay isn’t a video. It’s a recording of your movement positioned in the room, at the scale and location of the original action, viewed from wherever you choose to stand.
That changes three things at once:
- You can walk around your own mistake. Step to the side and watch the moment from an angle you couldn’t occupy while doing it. No camera placement decision is required, because the record is spatial rather than framed.
- The correction stays attached to the place. “Your hand was here, it should have been here” is a statement about two points in a room, and in AR it can be made as two points in that room rather than as words.
- The proprioceptive link survives. You’re standing where you stood, in the same body position, seconds later. The gap between the sensation and the feedback is small enough that they can be connected.
Fencing is a reasonable test case: fast, precise, heavily dependent on positioning and timing, and traditionally taught by a coach saying what went wrong immediately afterwards.
The wider application
Little about this is specific to fencing. Any skill taught by correcting movement has the same structure: dance, martial arts, instrumental technique, physical therapy, craft work at a bench, and the whole category of performance training.
It also sits in a line this site keeps returning to. Two days ago we covered LumiNote, a VR system for teaching stage lighting by turning an instructor’s spoken intent into spatial annotations. The common thread is that XR’s most credible near-term use in the arts is teaching, not performance: it’s the case where the technology’s actual strength — putting information in a specific place in space — lines up with a real pedagogical need, rather than competing with media that already work.
The research caveat applies as usual. This is a system paper with a study behind it, not a product, and “helps in an experiment” is a long way from “changes how anyone trains.” But the core idea — that feedback should live where the action was, not on a screen elsewhere — is worth taking into any interactive work that tries to teach someone something.
Related Reading
- FenceXR: AR Movement Replay for Error-Detection Training — arXiv:2609.19505
- arXiv cs.HC — recent human-computer interaction papers
- IEEE Computer Society — conferences
- LumiNote: LLM-Assisted Multimodal Instruction for VR Stage Lighting Education — arXiv:2609.17335
- ACM CHI — human factors in computing