On Foundation Models for Dynamical Systems from Purely Synthetic Data
This work addresses the problem of enabling foundation models for control applications, offering a novel approach to overcome data scarcity, though it is incremental in applying existing foundation model concepts to a new domain.
The paper tackles the lack of large-scale datasets for dynamical systems by pretraining a transformer-based foundation model exclusively on synthetic data, sampling dynamics functions from a reproducing kernel Hilbert space, and demonstrates that it outperforms specialist models in generalization, data efficiency, and robustness in tasks like cart-pole and Furuta pendulum setups.
Foundation models have demonstrated remarkable generalization, data efficiency, and robustness properties across various domains. In this paper, we explore the feasibility of foundation models for applications in the control domain. The success of these models is enabled by large-scale pretaining on Internet-scale datasets. These are available in fields like natural language processing and computer vision, but do not exist for dynamical systems. We address this challenge by pretraining a transformer-based foundation model exclusively on synthetic data and propose to sample dynamics functions from a reproducing kernel Hilbert space. Our pretrained model generalizes for prediction tasks across different dynamical systems, which we validate in simulation and hardware experiments, including cart-pole and Furuta pendulum setups. Additionally, the model can be fine-tuned effectively to new systems to increase performance even further. Our results demonstrate the feasibility of foundation models for dynamical systems that outperform specialist models in terms of generalization, data efficiency, and robustness.