PoSafeNet: Safe Learning with Poset-Structured Neural Nets

2026-01-29

AI Summary

Existing safe learning methods often struggle with heterogeneous safety constraints, as enforcing uniformity or fixed priorities can lead to infeasibility or brittle behavior, particularly in complex scenarios where only partial precedence relations exist among constraints. This work addresses this challenge by modeling safety constraints as a partially ordered set and introduces PoSafeNet—a differentiable neural safety layer that adaptively synthesizes valid safety policies through a closed-form sequential projection consistent with the underlying partial order. By allowing incomparable constraints to coexist while preserving priority semantics, PoSafeNet achieves substantial improvements in feasibility, robustness, and scalability. Empirical evaluations on multi-obstacle navigation, constrained robotic manipulation, and vision-based autonomous driving demonstrate its clear superiority over unstructured baselines and differentiable quadratic programming–based safety layers.

Technology Category

• Natural Language Processing: Safety and Robustness
• Philosophy and Ethics of AI: Safety, Robustness & Trustworthiness
• Search and Optimization: Learning to Search

Application Category

• Security and Privacy: Large-scale security measurements
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Safe learning is essential for deploying learningbased controllers in safety-critical robotic systems, yet existing approaches often enforce multiple safety constraints uniformly or via fixed priority orders, leading to infeasibility and brittle behavior. In practice, safety requirements are heterogeneous and admit only partial priority relations, where some constraints are comparable while others are inherently incomparable. We formalize this setting as poset-structured safety, modeling safety constraints as a partially ordered set and treating safety composition as a structural property of the policy class. Building on this formulation, we propose PoSafeNet, a differentiable neural safety layer that enforces safety via sequential closed-form projection under poset-consistent constraint orderings, enabling adaptive selection or mixing of valid safety executions while preserving priority semantics by construction. Experiments on multi-obstacle navigation, constrained robot manipulation, and vision-based autonomous driving demonstrate improved feasibility, robustness, and scalability over unstructured and differentiable quadratic program-based safety layers.

Problem

• safe learning
• safety constraints
• partial order
• robotic systems
• priority relations

Innovation

• poset-structured safety
• differentiable safety layer
• closed-form projection
• adaptive safety execution
• neural control