Bayesian Inverse Physics for Neuro-Symbolic Robot Learning

Octavio Arriaga, Rebecca Carrie Adam, Melvin Laux, Lisa Gutzeit, Marco Ragni, Jan Peters, Frank Kirchner
Proceedings of The 19th International Conference on Neurosymbolic Learning and Reasoning, PMLR 284:854-872, 2025.

Abstract

Real-world robotic applications, from autonomous exploration to assistive technologies, require adaptive, interpretable, and data-efficient learning paradigms. While deep learning architectures and foundation models have driven significant advances in diverse robotic applications, they remain limited in their ability to operate efficiently and reliably in unknown and dynamic environments. In this position paper, we critically assess these limitations and introduce a conceptual framework for combining data-driven learning with deliberate, structured reasoning. Specifically, we propose leveraging differentiable physics for efficient world modeling, Bayesian inference for uncertainty-aware decision-making, and meta-learning for rapid adaptation to new tasks. By embedding physical symbolic reasoning within neural models, robots could generalize beyond their training data, reason about novel situations, and continuously expand their knowledge. We argue that such hybrid neuro-symbolic architectures are essential for the next generation of autonomous systems, and to this end, we provide a research roadmap to guide and accelerate their development.

Cite this Paper

BibTeX
@InProceedings{pmlr-v284-arriaga25a, title = {Bayesian Inverse Physics for Neuro-Symbolic Robot Learning}, author = {Arriaga, Octavio and Adam, Rebecca Carrie and Laux, Melvin and Gutzeit, Lisa and Ragni, Marco and Peters, Jan and Kirchner, Frank}, booktitle = {Proceedings of The 19th International Conference on Neurosymbolic Learning and Reasoning}, pages = {854--872}, year = {2025}, editor = {H. Gilpin, Leilani and Giunchiglia, Eleonora and Hitzler, Pascal and van Krieken, Emile}, volume = {284}, series = {Proceedings of Machine Learning Research}, month = {08--10 Sep}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v284/main/assets/arriaga25a/arriaga25a.pdf}, url = {https://proceedings.mlr.press/v284/arriaga25a.html}, abstract = {Real-world robotic applications, from autonomous exploration to assistive technologies, require adaptive, interpretable, and data-efficient learning paradigms. While deep learning architectures and foundation models have driven significant advances in diverse robotic applications, they remain limited in their ability to operate efficiently and reliably in unknown and dynamic environments. In this position paper, we critically assess these limitations and introduce a conceptual framework for combining data-driven learning with deliberate, structured reasoning. Specifically, we propose leveraging differentiable physics for efficient world modeling, Bayesian inference for uncertainty-aware decision-making, and meta-learning for rapid adaptation to new tasks. By embedding physical symbolic reasoning within neural models, robots could generalize beyond their training data, reason about novel situations, and continuously expand their knowledge. We argue that such hybrid neuro-symbolic architectures are essential for the next generation of autonomous systems, and to this end, we provide a research roadmap to guide and accelerate their development.} }
Endnote
%0 Conference Paper %T Bayesian Inverse Physics for Neuro-Symbolic Robot Learning %A Octavio Arriaga %A Rebecca Carrie Adam %A Melvin Laux %A Lisa Gutzeit %A Marco Ragni %A Jan Peters %A Frank Kirchner %B Proceedings of The 19th International Conference on Neurosymbolic Learning and Reasoning %C Proceedings of Machine Learning Research %D 2025 %E Leilani H. Gilpin %E Eleonora Giunchiglia %E Pascal Hitzler %E Emile van Krieken %F pmlr-v284-arriaga25a %I PMLR %P 854--872 %U https://proceedings.mlr.press/v284/arriaga25a.html %V 284 %X Real-world robotic applications, from autonomous exploration to assistive technologies, require adaptive, interpretable, and data-efficient learning paradigms. While deep learning architectures and foundation models have driven significant advances in diverse robotic applications, they remain limited in their ability to operate efficiently and reliably in unknown and dynamic environments. In this position paper, we critically assess these limitations and introduce a conceptual framework for combining data-driven learning with deliberate, structured reasoning. Specifically, we propose leveraging differentiable physics for efficient world modeling, Bayesian inference for uncertainty-aware decision-making, and meta-learning for rapid adaptation to new tasks. By embedding physical symbolic reasoning within neural models, robots could generalize beyond their training data, reason about novel situations, and continuously expand their knowledge. We argue that such hybrid neuro-symbolic architectures are essential for the next generation of autonomous systems, and to this end, we provide a research roadmap to guide and accelerate their development.
APA
Arriaga, O., Adam, R.C., Laux, M., Gutzeit, L., Ragni, M., Peters, J. & Kirchner, F.. (2025). Bayesian Inverse Physics for Neuro-Symbolic Robot Learning. Proceedings of The 19th International Conference on Neurosymbolic Learning and Reasoning, in Proceedings of Machine Learning Research 284:854-872 Available from https://proceedings.mlr.press/v284/arriaga25a.html.

Related Material