Machine learning researchers build a better world model
Leibniz lives! Insight about how 'nature makes no leaps' promises better robot predictions
Machine learning researchers have developed a technique that improves world models, which are used to train physical AI systems like robots to navigate the real world. These models make predictions based on observed data and could provide better direction to robots if they could make inferences about possible actions without being trained for all outcomes.
The new approach, called Counterfactual World Models, can predict alternative scenarios by inferring possibilities from what actually happened. The technique uses the mismatch between predictions and reality to determine what went wrong and improve future predictions, debugging and reasoning at test time. Counterfactual world models can help robots retry actions after unexpected events, adapt to persistent faults like miscalibrated motors, and help people understand failures.
Written by urgent.news from The Register Science's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
