Generalizing Safety Beyond Collision-Avoidance via Latent-Space Reachability Analysis

2025-02-02

AI Summary

Real-world robotic safety risks—such as item spillage from bags or stack toppling—are often non-collisional and difficult to model explicitly. Method: This paper introduces a Hamilton–Jacobi (HJ) reachability analysis framework grounded in the latent space of generative world models (e.g., VAEs or diffusion models), enabling direct application of HJ theory to high-dimensional raw observations (e.g., RGB images). Safety boundary learning is reformulated as a classification task in the low-dimensional latent space, eliminating reliance on hand-crafted state representations and precise dynamical models. The approach integrates numerical solutions of the HJ partial differential equation with real-time safety-filtering control. Contribution/Results: The method provides plug-and-play safety guarantees for arbitrary policies—including generative and teleoperated ones. Evaluated in simulation and on a Franka Research 3 robot, it significantly suppresses failure modes such as toppling, spillage, and overturning, thereby enhancing safety and generalization in open-world environments.

Technology Category

• Natural Language Processing: Safety and Robustness
• Intelligent Robots: State Estimation
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Security and Privacy: Large-scale security measurements

Abstract

Hamilton-Jacobi (HJ) reachability is a rigorous mathematical framework that enables robots to simultaneously detect unsafe states and generate actions that prevent future failures. While in theory, HJ reachability can synthesize safe controllers for nonlinear systems and nonconvex constraints, in practice, it has been limited to hand-engineered collision-avoidance constraints modeled via low-dimensional state-space representations and first-principles dynamics. In this work, our goal is to generalize safe robot controllers to prevent failures that are hard -- if not impossible -- to write down by hand, but can be intuitively identified from high-dimensional observations: for example, spilling the contents of a bag. We propose Latent Safety Filters, a latent-space generalization of HJ reachability that tractably operates directly on raw observation data (e.g., RGB images) by performing safety analysis in the latent embedding space of a generative world model. This transforms nuanced constraint specification to a classification problem in latent space and enables reasoning about dynamical consequences that are hard to simulate. In simulation and hardware experiments, we use Latent Safety Filters to safeguard arbitrary policies (from generative policies to direct teleoperation) from complex safety hazards, like preventing a Franka Research 3 manipulator from spilling the contents of a bag or toppling cluttered objects.

Problem

• Robot Safety
• Complex Hazard Avoidance
• Real-world Practicality

Innovation

• Latent Safety Filter
• Complex Obstacle Avoidance
• Robotic Safety Enhancement