Benchmarking Generative Trajectory Models for Active-Inference Control

2026-10-05

AI Summary

This study addresses the challenge that the complex dynamics of real-world systems are difficult to model explicitly, which limits the application of active inference in control. We propose GenAIF, a framework that leverages generative trajectory models to learn policy distributions and observation likelihood mappings from demonstrations. By unifying action proposal, controlled prediction, and observational evidence within a shared generative model, GenAIF enables active inference-based control. We systematically evaluate diffusion models, autoregressive Transformers, conditional variational autoencoders (CVAEs), and flow matching techniques on MuJoCo benchmarks. Our results demonstrate that diffusion models achieve superior control performance, while CVAEs offer higher inference efficiency. Furthermore, this work validates the critical role of proper conditioning in facilitating effective belief adaptation.

Abstract

Learning from trajectory demonstrations offers a route to active-inference control of complex systems whose dynamics are difficult to model explicitly. We introduce generative active-inference control (GenAIF), in which one generative trajectory model learns from demonstrations and measured action interventions to supply a goal-conditioned policy distribution and a state-to-observation likelihood mapping. From this control design, we derive three model requirements: (i) useful action proposals, (ii) accurate prediction under imposed actions, and (iii) probabilistic observation evidence for belief updating and expected information gain. We benchmark diffusion, autoregressive Transformers, conditional variational autoencoders (CVAEs), and flow matching in a MuJoCo manipulation task with multiple physical conditions. Diffusion delivers the strongest control across the tested dynamics, while CVAE combines comparable short-horizon prediction with much faster inference. Correct conditioning is decisive, and trajectory reuse offers further computational savings. With the same frozen models, a hidden-dynamics experiment demonstrates prompt belief adaptation after an unannounced tilt change; subsequent instability identifies sustained inference as a remaining challenge. These findings support the use of shared generative trajectory models to connect action proposal, controlled prediction, and observation evidence within GenAIF.

Problem

• active-inference control
• generative trajectory models
• benchmarking
• complex systems
• learning from demonstrations

Innovation

• Generative Active-Inference Control
• Trajectory Models
• Diffusion Models
• Conditional Variational Autoencoders
• Belief Updating