I observed that robots remain clumsy because of three core uncertainties: Control, Physics, and Perception. I also noted that Labellerr AI expects robots to become useful and safer in homes within the next 10 to 20 years.
- Uncertainty in Control
I anticipate domestic robots will become useful and safer within the next 10 to 20 years.
- Uncertainty in Physics
- Uncertainty in Perception
"photons in and controls out"
I use the phrase “photons in and controls out” to highlight how both machines and humans ingest visual information (photons) and emit control signals to act, unifying self‑driving cars, humanoid robots, and humans under a single information‑flow metaphor. See the video for a deeper dive.
I concluded that a robot waiter needs a World Model plus Model Predictive Control (MPC) rather than relying solely on YOLO detection, because YOLO‑only systems produce jerky, scared motions.
- World Model creates a short‑term predictive movie of the scene
- MPC evaluates ~1,000 candidate trajectories per cycle and selects the smoothest path
- VLA/VLM translates high‑level commands into planner goals