From Prediction to Consequence: A New Evaluation Paradigm for Embodied AI

Talk overview

Modern AI excels at predicting actions from offline data, but embodied systems require evaluating how predictions affect real-world outcomes. This talk introduces a new evaluation framework that combines open-loop and closed-loop methods to assess prediction-to-consequence alignment. It uses autonomous driving and AlpaBridge, an external-driver interface, to demonstrate how policies behave differently when actions influence environment dynamics.