Tobias Pfaff

Tobias Pfaff

Auto-generated profile, based on publicly available information

Tobias Pfaff

Google DeepMind, London N1C 4DN, UK

Research Focus & Metrics

Academic Career Length

14Years (2013-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
17 September 2025 in Science Robotics
Verified
Journal Article
Closed access
BABo Ai
Bo Ai
STStephen Tian
Stephen Tian
HSHaochen Shi
Haochen Shi
YWYixuan Wang
Yixuan Wang
TPTobias Pfaff
Tobias Pfaff
YLYunzhu Li
Yunzhu Li
Dynamics models that predict the effects of physical interactions are essential for planning and control in robotic manipulation. Although models based on physical principles often generalize well, they typically require full-state information, which can be difficult or impossible to extract from perception data in complex, real-world scenarios. Learning-based dynamics models provide an alternative by deriving state transition functions purely from perceived interaction data, enabling the capture of complex, hard-to-model factors and predictive uncertainty and accelerating simulations that are often too slow for real-time control. Recent successes in this field have demonstrated notable advancements in robot capabilities, including long-horizon manipulation of deformable objects, granular materials, and complex multiobject interactions such as stowing and packing. A crucial aspect of these investigations is the choice of state representation, which determines the inductive biases in the learning system for reduced-order modeling of scene dynamics. This article provides a timely and comprehensive review of current techniques and trade-offs in designing learned dynamics models, highlighting their role in advancing robot capabilities through integration with state estimation and control and identifying critical research gaps for future exploration.

21 July 2013
Verified
Conference Paper
Open access
NTNils Thuerey
Nils Thuerey
TKTheodore Kim
Theodore Kim
TPTobias Pfaff
Tobias Pfaff
Over the last decade, the visual effects industry has embraced physics simulations as a highly useful tool for creating realistic scenes ranging from a small camp fire to large-scale destruction of whole cities. While fluid simulations are now widely used in the industry, it is still inherently difficult to control large-scale simulations, and there is a constant struggle to increase visual detail. This course approaches these problems using turbulence methods. Turbulent detail is what makes typical fluid simulations look impressive, and the underlying physics motivate a powerful approach for control; they allow for an elegant split of large-scale motion and small-scale turbulent detail. The result is a two-stage work flow that is highly convenient for artists. First, a rough, fast initial simulation is performed, then turbulent effects are added to enhance detail. After reviewing the basics of fluid solvers and the popular wavelet turbulence approach, the course presents several powerful methods for capturing advanced effects such as boundary layers and turbulence with directional preferences. It also explains the difficulties of liquid simulations and presents an approach to liquid turbulence that is based on wave dynamics. Full source code for all of the methods covered in the course is available to attendees. Instructors outline convenient starting points for navigating the code.

1 February 2022 in arXiv
Verified
Preprint
Open access
PDF file
KAKelsey R. Allen
Kelsey R. Allen
TLTatiana Lopez-Guevara
Tatiana Lopez-Guevara
KSKimberly Stachenfeld
Kimberly Stachenfeld
ASAlvaro Sanchez-Gonzalez
Alvaro Sanchez-Gonzalez
PBPeter Battaglia
Peter Battaglia
TPTobias Pfaff
Tobias Pfaff
Designing physical artifacts that serve a purpose - such as tools and other functional structures - is central to engineering as well as everyday human behavior. Though automating design has tremendous promise, general-purpose methods do not yet exist. Here we explore a simple, fast, and robust approach to inverse design which combines learned forward simulators based on graph neural networks with gradient-based design optimization. Our approach solves high-dimensional problems with complex physical dynamics, including designing surfaces and tools to manipulate fluid flows and optimizing the shape of an airfoil to minimize drag. This framework produces high-quality designs by propagating gradients through trajectories of hundreds of steps, even when using models that were pre-trained for single-step predictions on data substantially different from the design tasks. In our fluid manipulation tasks, the resulting designs outperformed those found by sampling-based optimization techniques. In airfoil design, they matched the quality of those obtained with a specialized solver. Our results suggest that despite some remaining challenges, machine learning-based simulators are maturing to the point where they can support general-purpose design optimization across a variety of domains.

9 May 2026 in Results in Engineering
Verified
Journal Article
Open access
HLHaoran Li
Haoran Li
YZYingxue Zhao
Yingxue Zhao
HZHaosu Zhou
Haosu Zhou
TPTobias Pfaff
Tobias Pfaff
NLNan Li
Nan Li