Synthesizing control policies that preserve the safety of autonomous systems is a challenge that remains to be solved. Towards that goal, control barrier functions (CBFs) have been developed as mathematical constructs that can be used in real-time to correct safety-violating nominal actions to ones which preserve the safety of control systems. However, synthesizing CBFs using correct-by-construction methods has not been scalable. Instead, recent research has proposed data-driven approaches for learning CBFs in the form of neural networks. Two main challenges face such approaches: (1) labeling states as unsafe or safe ones requires the knowledge of the states in the backward reachable set of the failure set--the true dynamics-dependent unsafe set, and (2) in the case of systems with high-dimensional observations, such as images and point clouds, enormous amount of data is needed to train these neural observation-based CBFs, which is expensive to obtain in robotic domains. We tackle the first challenge by using inverse constraint learning to infer a neural classifier that defines the backward reachable set from expert trajectories and use it to label sampled states. This method outperforms baselines and performs comparably to a CBF trained with ground truth labels in four environments. We tackle the second challenge by using existing vision models which are pre-trained on large and diverse datasets as frozen perception backbones on top of which latent dynamics and neural observation-based CBFs are trained. Our experimental results indicate that the resulting filters are competitive with those that have access to the ground truth state.
","abstract_html":"<p>Synthesizing control policies that preserve the safety of autonomous systems is a challenge that remains to be solved. Towards that goal, control barrier functions (CBFs) have been developed as mathematical constructs that can be used in real-time to correct safety-violating nominal actions to ones which preserve the safety of control systems. However, synthesizing CBFs using correct-by-construction methods has not been scalable. Instead, recent research has proposed data-driven approaches for learning CBFs in the form of neural networks. Two main challenges face such approaches: (1) labeling states as unsafe or safe ones requires the knowledge of the states in the backward reachable set of the failure set--the true dynamics-dependent unsafe set, and (2) in the case of systems with high-dimensional observations, such as images and point clouds, enormous amount of data is needed to train these neural observation-based CBFs, which is expensive to obtain in robotic domains. We tackle the first challenge by using inverse constraint learning to infer a neural classifier that defines the backward reachable set from expert trajectories and use it to label sampled states. This method outperforms baselines and performs comparably to a CBF trained with ground truth labels in four environments. We tackle the second challenge by using existing vision models which are pre-trained on large and diverse datasets as frozen perception backbones on top of which latent dynamics and neural observation-based CBFs are trained. Our experimental results indicate that the resulting filters are competitive with those that have access to the ground truth state.</p>","abstract_has_math":false,"creators":["Yang, Yuxuan"],"institution":null,"degree_name":"Master of Science (MS)","degree_level":"Thesis","degree_discipline":"Computer Science & Engineering","degree_department":null,"school":null,"contributors":["Hussein Sibai","Andrew Clark, Nathan Jacobs"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025-05-07T07:00:00Z","date_published":"2025-05-07T07:00:00Z","updated_at":"2026-07-24T06:13:55Z","subjects":["Control Barrier Functions","Safety Filters","Inverse Constraint Learning","Computer Vision","Engineering","Navigation, Guidance, Control, and Dynamics"],"languages":["English (en)"],"rights":[],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["https://openscholarship.wustl.edu/eng_etds/1217"],"render_values":[{"text":"https://openscholarship.wustl.edu/eng_etds/1217","href":"https://openscholarship.wustl.edu/eng_etds/1217","code":true}]}]},"links":{"outbound_url":"https://doi.org/10.7936/r36a-2b25","outbound_label":"DOI","outbound_source":"dc:identifier"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Hussein Sibai","Andrew Clark, Nathan Jacobs"]},{"key":"dc:creator","label":"Author","values":["Yang, Yuxuan"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.available","label":"Dc Date Available","values":["2025-05-07T07:00:00Z"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Computer Science & Engineering","McKelvey School of Engineering"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Thesis"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Master of Science (MS)"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Control Barrier Functions","Safety Filters","Inverse Constraint Learning","Computer Vision","Engineering","Navigation, Guidance, Control, and Dynamics"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["English (en)"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://doi.org/10.7936/r36a-2b25","https://openscholarship.wustl.edu/eng_etds/1217"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Synthesizing control policies that preserve the safety of autonomous systems is a challenge that remains to be solved. Towards that goal, control barrier functions (CBFs) have been developed as mathematical constructs that can be used in real-time to correct safety-violating nominal actions to ones which preserve the safety of control systems. However, synthesizing CBFs using correct-by-construction methods has not been scalable. Instead, recent research has proposed data-driven approaches for learning CBFs in the form of neural networks. Two main challenges face such approaches: (1) labeling states as unsafe or safe ones requires the knowledge of the states in the backward reachable set of the failure set--the true dynamics-dependent unsafe set, and (2) in the case of systems with high-dimensional observations, such as images and point clouds, enormous amount of data is needed to train these neural observation-based CBFs, which is expensive to obtain in robotic domains. We tackle the first challenge by using inverse constraint learning to infer a neural classifier that defines the backward reachable set from expert trajectories and use it to label sampled states. This method outperforms baselines and performs comparably to a CBF trained with ground truth labels in four environments. We tackle the second challenge by using existing vision models which are pre-trained on large and diverse datasets as frozen perception backbones on top of which latent dynamics and neural observation-based CBFs are trained. Our experimental results indicate that the resulting filters are competitive with those that have access to the ground truth state.
"]},{"key":"dc:title","label":"Title","values":["Training Safety Control Filters Using High-dimensional and Un-labeled Data"]}]}],"canonical_facts":{"dc:contributor":["Hussein Sibai","Andrew Clark, Nathan Jacobs"],"dc:creator":["Yang, Yuxuan"],"dc:date.available":["2025-05-07T07:00:00Z"],"dc:description.abstract":["Synthesizing control policies that preserve the safety of autonomous systems is a challenge that remains to be solved. Towards that goal, control barrier functions (CBFs) have been developed as mathematical constructs that can be used in real-time to correct safety-violating nominal actions to ones which preserve the safety of control systems. However, synthesizing CBFs using correct-by-construction methods has not been scalable. Instead, recent research has proposed data-driven approaches for learning CBFs in the form of neural networks. Two main challenges face such approaches: (1) labeling states as unsafe or safe ones requires the knowledge of the states in the backward reachable set of the failure set--the true dynamics-dependent unsafe set, and (2) in the case of systems with high-dimensional observations, such as images and point clouds, enormous amount of data is needed to train these neural observation-based CBFs, which is expensive to obtain in robotic domains. We tackle the first challenge by using inverse constraint learning to infer a neural classifier that defines the backward reachable set from expert trajectories and use it to label sampled states. This method outperforms baselines and performs comparably to a CBF trained with ground truth labels in four environments. We tackle the second challenge by using existing vision models which are pre-trained on large and diverse datasets as frozen perception backbones on top of which latent dynamics and neural observation-based CBFs are trained. Our experimental results indicate that the resulting filters are competitive with those that have access to the ground truth state.
"],"dc:identifier":["https://doi.org/10.7936/r36a-2b25","https://openscholarship.wustl.edu/eng_etds/1217"],"dc:language":["English (en)"],"dc:subject":["Control Barrier Functions","Safety Filters","Inverse Constraint Learning","Computer Vision","Engineering","Navigation, Guidance, Control, and Dynamics"],"dc:title":["Training Safety Control Filters Using High-dimensional and Un-labeled Data"],"thesis:degree_discipline":["Computer Science & Engineering","McKelvey School of Engineering"],"thesis:degree_level":["Thesis"],"thesis:degree_name":["Master of Science (MS)"]},"updated_at":"2026-07-24T06:13:55Z"}