Model posterior updates and subsequent control decisions
Develop an explicit model of how posterior updates alter subsequent control decisions in task-oriented information acquisition for model predictive path integral control, extending the computationally tractable surrogate toward Bayesian value-of-information planning.
References
Explicitly modeling how posterior updates alter subsequent control decisions is left for future work.
— Task-Oriented Active Learning of Residual Dynamics for Model Predictive Path Integral Control
(2609.19378 - Aoki et al., 16 Sep 2026) in Section 6, Discussion
At the tested settings, diffusion gives the strongest direct control, while CVAE provides similar short-horizon position prediction at lower cost. Reuse reduces computation with model-dependent effects on control. All four models initially identify the dynamics change but fail to sustain confidence; belief-guided action selection remains open.
— Benchmarking Generative Trajectory Models for Active-Inference Control
(2610.05692 - Li et al., 5 Oct 2026) in Section 1, Introduction