CrossSafe: Towards Cross-Embodiment Latent Safety Filters
A morphology-aware latent Hamilton–Jacobi safety filter that transfers across bimanual robot embodiments, including robots held out of training.
Washington University in St. Louis
Trustworthy Autonomy Lab
We develop rigorous, scalable methods for designing, verifying, and deploying safe autonomous systems at the intersection of formal methods, control theory, robotics, and machine learning.
Selected publications
Research project pages
Project pages collect each paper’s abstract, results, resources, and citation.
Safe Control using Learned Safety Filters and Adaptive Conformal Inference
Adaptive Conformal Filtering calibrates learned Hamilton–Jacobi safety filters and provides statistical guarantees.
Towards General Language-Conditioned Latent Safety Filters
A language-conditioned Hamilton–Jacobi safety actor and critic for enforcing varying safety requirements in robotic manipulation.
Statistically Assuring Safety of Control Systems using Ensembles of Safety Filters and Conformal Prediction
Conformal prediction provides probabilistic safety guarantees for learned Hamilton–Jacobi safety filters and ensembles.
Computing Sound Lower and Upper Bounds on Hamilton-Jacobi Reach-Avoid Value Functions
Sound lower and upper bounds account for grid discretization errors in backward reachable and reach-avoid sets.
Learning Neural Control Barrier Functions from Expert Demonstrations using Inverse Constraint Learning
Learning neural control barrier functions from expert demonstrations when the failure set and ground-truth safety labels are unavailable.
Designing Latent Safety Filters using Pre-Trained Vision Models
Evaluating pre-trained vision representations as perception backbones for vision-based safety filters.
Learning Conservative Neural Control Barrier Functions from Offline Data
Training conservative neural control barrier functions that avoid unsafe and out-of-distribution states using offline data.