GenEA is a novel agent representation that can reason about multi‑step, long‑horizon embodied plans. Its core is a multilevel state‑space that couples coarse biomechanics of an articulated body with interactions with nearby objects, extending the task‑and‑motion planning framework with simulation‑based intuitive physics. By embedding this state‑space in a generative model, GenEA yields a joint probability distribution over goals and embodied plans. Across experiments in object manipulation and navigation covering diverse tasks and behavioral measures, simple transformations of this joint distribution allow GenEA to recapitulate how humans reason about embodied plans. Compared with plausible heuristics, standard task‑and‑motion planning, and a state‑of‑the‑art vision‑language model, GenEA consistently achieves higher prediction accuracy and better behavioral alignment. This suggests that human reasoning and social cognition may fundamentally rely on generative action‑planning models, and embodied AI systems aiming for seamless human interaction should adopt GenEA‑like generative planning architectures.
Review: GenEA demonstrates the power of integrating physical simulation with generative modeling, offering a promising direction for interpretable reasoning in embodied AI.