Pixels to Proofs: Probabilistically-Safe Latent World Model Control via Parallel Conformal Robust MPC

The article introduces SLS², a framework for safe feedback motion planning utilizing robust model predictive control (MPC) with learned latent world models. It features a compact Markovian latent state representation and employs GPU-accelerated system-level synthesis to incorporate conformal prediction for calibrated error bounds, enhancing safety during trajectory optimization. The method demonstrates improved goal-reaching performance and safety in vision-based control tasks, making it a significant advancement for practitioners focused on integrating safety in AI-driven motion planning.