Show HN: AI climbing coach – visualize how to climb any route based on your body

Original Article ↗ Hacker News Discussion ↗

Technical approach & capabilities

  • System builds a 3D avatar of the climber from the first second of video and runs an end‑to‑end biomechanics model from video frames.
  • It infers camera parameters and, apparently, some monocular depth to reason about the wall, though accuracy on complex 3D cliffs is limited.
  • Path planning up the wall and generation of movement/kinematics are integrated into the same pipeline.
  • Works best on indoor bouldering with clear, spaced holds and good lighting; outdoor and natural walls are possible but less reliable.
  • Hold identification and wall understanding are implicitly learned rather than explicitly modeled; exact mechanisms are not fully described.

Open source, costs, and maturity

  • Author plans to open source code, training pipeline, and model, but notes heavy compute and data requirements for high‑quality results.
  • Current system is an early research preview: works well on some examples, fails on many; demo clips are cherry‑picked.
  • Licensing constraints around certain body models (e.g., SMPL‑family) were raised; commercial use may be restricted.

Use cases and perceived value

  • Suggested uses: beginner coaching, visual beta suggestions, comparing your attempts to an “optimal” avatar, movement analysis, gym features (ghost climber overlays), and possible support for route setters.
  • Some see parallels to golf swing or tape analysis in other sports.

Limitations & skepticism

  • Many climbers doubt it can handle micro‑details: hold texture, polish, tiny orientation changes, dual textures, and subtle friction differences.
  • Concerns that it ignores climber‑specific strength, flexibility, and injury constraints, so suggested beta may be physically unattainable.
  • Questions about how it knows which holds are “on,” where the top is, and how it infers 3D positions of unused holds; these remain unclear.
  • Some view route grading or safety‑critical advice as unreliable without texture/depth and worry about trusting AI in dangerous contexts.

Climbing culture reactions

  • Strong split: some are excited and eager to try it; many say it removes a core joy of climbing—solving the puzzle themselves and sharing beta with partners.
  • Others argue structured feedback and expert‑like beta can drastically speed learning, especially for newer climbers.
  • Broader debate appears about AI’s role in creative tasks like route setting and whether algorithmic routes could or should replace human craft.