Information-Aware Model Predictive Control for Satellite Inspection
arXiv:2608.07765
2026
Dynamics
2 ideas extracted · analyzed Sep 1, 2026
What the math gives to ML
The paper embeds Kalman-filter covariance propagation inside an MPC optimization, so the controller chooses actions that satisfy dynamics and safety constraints while improving future measurement geometry. Its transferable mechanism is an information-aware rollout objective: actions are evaluated not only by task cost but also by the covariance they induce through state-dependent observation models. This can be transferred to neural world models, active-sensing policies, and attention-based agents by optimizing predicted latent uncertainty reduction jointly with task performance.
Ideas from this paper
✓✓ Beats tuned baseline
2026
Add a differentiable uncertainty state to a learned world model and optimize action sequences using both predicted task reward and the covariance of the latent or target-state estimator. The policy should move or attend toward states that make observations informative, rather than selecting actions only from mean-state predictions.
Useful8/10
Difficulty6/10
Novelty6/10
✗ Failed on benchmark
2026
Use predicted covariance reduction as a differentiable gate for selecting tokens, views, sensors, or retrieved demonstrations. The gate favors inputs with high expected information gain while accounting for acquisition cost, turning attention and data collection into active observability optimization.
Useful7/10
Difficulty5/10
Novelty7/10