aiqbench analysis
This research proposes a hybrid framework combining an action-conditioned JEPA world model with deterministic safety shields to enable safe execution for heterogeneous robots without relying solely on probabilistic guarantees from large policies. By using the predictive architecture only to rank candidate actions generated by classical planners, the system maintains rigorous kinematic constraints while improving task success rates in simulation environments like LIBERO-Long. The approach addresses a critical gap where generalizable AI models lack real-time execution safety, offering a pathway toward deploying advanced robotics on edge accelerators with verified risk mitigation.
Calibrated Predictive Safety for Heterogeneous Robots: An Action-Conditioned JEPA Framework with Model-Based Safety Shields
arXiv:2608.17496v1 Announce Type: new Abstract: Vision-language-action policies generalize broadly but provide no execution-time guarantees; classical model-based planners respect kinematic and geometric constraints but generalize poorly. We study whether an action-conditioned Joint-Embedding Predictive Architecture (JEPA) world model can predict, before execution, both task progress and physical risk for candidate action chunks, and whether coupling these predictions to an embodiment-specific model-based safety shield yields a deployable pipeline for heterogeneous robots. We propose a receding-horizon decision pipeline: (1) a proposer produces K candidate action chunks; (2) an action-conditioned JEPA rolls each candidate forward in a frozen-encoder latent space conditioned on an embodiment embedding; (3) calibrated risk and progress heads score each rollout and report uncertainty; (4) a deterministic per-embodiment safety shield filters inadmissible candidates; (5) a fallback ladder handles empty-admissible-set cases. The learned ranking only reorders admissible candidates; enforcement guarantees come from the deterministic shield and fallback ladder. We evaluate with a pre-registered protocol in simulation (LIBERO-Long). In 600-episode configurations the full framework improved success over a shield-only baseline and reduced collision false negatives at matched recall. Deployment-efficiency measurements on target on-robot and edge accelerators are included. Real-robot experiments and an offline reranking significance test remain future work; see the paper for disclosures.
Abstract:Vision-language-action policies generalize broadly but provide no execution-time guarantees; classical model-based planners respect kinematic and geometric constraints but generalize poorly. We study whether an action-conditioned Joint-Embedding Predictive Architecture (JEPA) world model can predict, before execution, both task progress and physical risk for candidate action chunks, and whether coupling these predictions to an embodiment-specific model-based safety shield yields a deployable pipeline for heterogeneous robots. We propose a receding-horizon decision pipeline: (1) a proposer produces K candidate action chunks; (2) an action-conditioned JEPA rolls each candidate forward in a frozen-encoder latent space conditioned on an embodiment embedding; (3) calibrated risk and progress heads score each rollout and report uncertainty; (4) a deterministic per-embodiment safety shield filters inadmissible candidates; (5) a fallback ladder handles empty-admissible-set cases. The learned ranking only reorders admissible candidates; enforcement guarantees come from the deterministic shield and fallback ladder. We evaluate with a pre-registered protocol in simulation (LIBERO-Long). In 600-episode configurations the full framework improved success over a shield-only baseline and reduced collision false negatives at matched recall. Deployment-efficiency measurements on target on-robot and edge accelerators are included. Real-robot experiments and an offline reranking significance test remain future work; see the paper for disclosures.
17 pages, 9 figures. Simulation-only empirical results on LIBERO-Long (no real-robot experiments). Source, figure-generation scripts and reproducibility checklist included. Level-3 offline reranking significance test not executed; see Sec. 7 (Scope and honesty statement) for detailed disclosure
Robotics (cs.RO)
arXiv:2608.17496 [cs.RO]
(or arXiv:2608.17496v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2608.17496 arXiv-issued DOI via DataCite (pending registration)
Full text under the publisher's open licence · 248 words
Related coverage
Uncovering Vulnerability of Vision-Language-Action Models under Joint-Level Physical Faults
arXiv · Jun 10, 2026
TS-WAM: Temporal Latent Actions and World-Action Coupling for Vision-Language Robot Control
arXiv · Oct 16, 2026
The J-Space Debate, Agent Swarms, and Pacing Frontier AI - Digital Minds Newsletter #4
LessWrong · Sep 18, 2026
Beyond Safe Execution: Safe Online Reinforcement Learning for Autonomous Driving with Value Calibration and Adaptive Guidance
arXiv · Oct 14, 2026