See the Wiki page for more info.

Robotic AI covers intelligent systems that perceive, model, and act in the physical world. Topics covered include:

  • JEPA (Joint Embedding Predictive Architecture) — world models and self-supervised learning
  • Belief tracking — state estimation, sensor fusion, Bayesian filtering
  • Control loops — PID, model predictive control, feedback systems
  • Autonomy — decision-making, path planning, mission execution

I took special interest in robotics AI when I first started to read about JEPA and Yan LeCun’s future plans. The demos are all low-level simulation meant to demo the core concepts. I was surprised by LeCun’s claims that his future version of JEPA would be truly intelligent. This, along with the results of the JEPA demos, made me lose interest.

Drones and airliners can fly autonomously in the air because they’ve got so much room for error. But robots (cars/humanoids) have to operate in far most complex scenarios (complex road surfaces, people and animals, traffic, road weather conditions, etc). There is little room for error. Building and testing robotics is an expensive team effort (not something I can research with a team consisting of myself and ChatGPT). In any case, my interest in robotics AI is still strong.


roboticAI


26.0425