From Space to Time: Enabling Adaptive Safety with Learned Value Functions via Disturbance Recasting

Sander Tonkens, Nikhil Uday Shinde, Azra Begzadić, Michael C. Yip, Jorge Cortes, Sylvia Lee Herbert
Proceedings of The 9th Conference on Robot Learning, PMLR 305:4103-4122, 2025.

Abstract

Safe operation is essential for autonomous systems in safety-critical environments such as urban air mobility. Value function-based safety filters provide formal guarantees on safety, wrapping learned or planning-based controllers with a layer of protection. Recent approaches leverage offline learned value functions to scale these safety filters to high-dimensional systems. Yet these methods assume detailed prior knowledge of all possible sources of model mismatch, in the form of disturbances, in the environment – information that is typically unavailable in real world settings. Even in well-mapped environments like urban canyons or industrial sites, drones encounter complex, spatially-varying disturbances arising from payload-drone interaction, turbulent airflow, and other environmental factors. We introduce Space2Time, which enables safe and adaptive deployment of offline-learned safety filters under unknown, spatially-varying disturbances. The key idea is to reparameterize spatial disturbances as a time-varying formulation, allowing the use of temporally varying precomputed value functions during online operation. We validate Space2Time through extensive simulations on diverse quadcopter models and real-world hardware experiments, demonstrating significantly improved safety performance over worst-case and naive baselines.

Cite this Paper

BibTeX
@InProceedings{pmlr-v305-tonkens25a, title = {From Space to Time: Enabling Adaptive Safety with Learned Value Functions via Disturbance Recasting}, author = {Tonkens, Sander and Shinde, Nikhil Uday and Begzadi\'{c}, Azra and Yip, Michael C. and Cortes, Jorge and Herbert, Sylvia Lee}, booktitle = {Proceedings of The 9th Conference on Robot Learning}, pages = {4103--4122}, year = {2025}, editor = {Lim, Joseph and Song, Shuran and Park, Hae-Won}, volume = {305}, series = {Proceedings of Machine Learning Research}, month = {27--30 Sep}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v305/main/assets/tonkens25a/tonkens25a.pdf}, url = {https://proceedings.mlr.press/v305/tonkens25a.html}, abstract = {Safe operation is essential for autonomous systems in safety-critical environments such as urban air mobility. Value function-based safety filters provide formal guarantees on safety, wrapping learned or planning-based controllers with a layer of protection. Recent approaches leverage offline learned value functions to scale these safety filters to high-dimensional systems. Yet these methods assume detailed prior knowledge of all possible sources of model mismatch, in the form of disturbances, in the environment – information that is typically unavailable in real world settings. Even in well-mapped environments like urban canyons or industrial sites, drones encounter complex, spatially-varying disturbances arising from payload-drone interaction, turbulent airflow, and other environmental factors. We introduce Space2Time, which enables safe and adaptive deployment of offline-learned safety filters under unknown, spatially-varying disturbances. The key idea is to reparameterize spatial disturbances as a time-varying formulation, allowing the use of temporally varying precomputed value functions during online operation. We validate Space2Time through extensive simulations on diverse quadcopter models and real-world hardware experiments, demonstrating significantly improved safety performance over worst-case and naive baselines.} }
Endnote
%0 Conference Paper %T From Space to Time: Enabling Adaptive Safety with Learned Value Functions via Disturbance Recasting %A Sander Tonkens %A Nikhil Uday Shinde %A Azra Begzadić %A Michael C. Yip %A Jorge Cortes %A Sylvia Lee Herbert %B Proceedings of The 9th Conference on Robot Learning %C Proceedings of Machine Learning Research %D 2025 %E Joseph Lim %E Shuran Song %E Hae-Won Park %F pmlr-v305-tonkens25a %I PMLR %P 4103--4122 %U https://proceedings.mlr.press/v305/tonkens25a.html %V 305 %X Safe operation is essential for autonomous systems in safety-critical environments such as urban air mobility. Value function-based safety filters provide formal guarantees on safety, wrapping learned or planning-based controllers with a layer of protection. Recent approaches leverage offline learned value functions to scale these safety filters to high-dimensional systems. Yet these methods assume detailed prior knowledge of all possible sources of model mismatch, in the form of disturbances, in the environment – information that is typically unavailable in real world settings. Even in well-mapped environments like urban canyons or industrial sites, drones encounter complex, spatially-varying disturbances arising from payload-drone interaction, turbulent airflow, and other environmental factors. We introduce Space2Time, which enables safe and adaptive deployment of offline-learned safety filters under unknown, spatially-varying disturbances. The key idea is to reparameterize spatial disturbances as a time-varying formulation, allowing the use of temporally varying precomputed value functions during online operation. We validate Space2Time through extensive simulations on diverse quadcopter models and real-world hardware experiments, demonstrating significantly improved safety performance over worst-case and naive baselines.
APA
Tonkens, S., Shinde, N.U., Begzadić, A., Yip, M.C., Cortes, J. & Herbert, S.L.. (2025). From Space to Time: Enabling Adaptive Safety with Learned Value Functions via Disturbance Recasting. Proceedings of The 9th Conference on Robot Learning, in Proceedings of Machine Learning Research 305:4103-4122 Available from https://proceedings.mlr.press/v305/tonkens25a.html.

Related Material