New framework enables safe motion planning with latent world models

Researchers have developed SLS^2, a novel framework for safe motion planning that utilizes robust model predictive control (MPC) within learned latent world models. This approach trains an action-conditioned world model with compact latent states, allowing for efficient trajectory optimization. To ensure system safety despite prediction inaccuracies, the framework incorporates conformal prediction to establish calibrated latent error bounds and robust constraint sets, which are then used by a GPU-accelerated MPC scheme. Additionally, a learned and conformalized latent constraint checker is employed to enforce probabilistic safety during closed-loop execution, demonstrating improved goal-reaching performance and safety in vision-based control tasks compared to existing methods. AI

RANK_REASON The cluster contains a research paper published on arXiv detailing a new framework for safe motion planning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Gotit.pub
• graphics processing unit
• Hugging Face
• model predictive control
• ScienceCast
• SLS^2
• Super-Level-Set Regression

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →