Generalizing Safety Beyond Collision Avoidance via Latent-Space Reachability Analysis

Kensuke Nakamura, Lasse Peters, Andrea Bajcsy

Keywords: safe control, world models

TL;DR: We introduce a latent generalization of Hamilton-Jacobi reachability using pretrained world models to enable us to safeguard against hard-to-model constraints beyond collisions

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) to automatically computes safety-preserving actions
without explicit recovery demonstrations by performing safety
analysis in the latent embedding space of a generative world
model. Our method leverages diverse robot observation-action
data of varying quality to learn a world model. Constraint
specification is then transformed into a classification problem
in the latent space of the learned world model. In hardware
experiments, we use Latent Safety Filters to safeguard arbitrary
policies (from imitation-learned policies to direct teleoperation)
from complex safety hazards, like preventing a Franka Research
3 manipulator from spilling the contents of a bag

Submission Number: 1