Preston Culbertson

Postdoctoral Scholar,
California Institute of Technology

email | CV | twitter | google scholar | github

I'm a postdoc at Caltech, working in the AMBER Lab with Prof. Aaron Ames. Before Caltech, I completed my PhD at Stanford University, supervised by Prof. Mac Schwager.

I am currently on the academic job market!

During my PhD, I also had the pleasure to spend my summers with the multi-robot group at JPL.

Outside of research, I love to bake, rock climb, volunteer, and wrangle my grey tabby, Huxley.

Picture of Preston Culbertson.


I am curious about how robots reason about uncertainty in their environment, whether it comes from inaccurate models of the world or their dynamics, uncertain state estimates, or the actions of other agents.

My aim is to use tools from nonlinear control, Bayesian estimation, and modern, learning-based computer vision to build robots that can properly estimate their risk online, and respond accordingly. I am excited particularly by applications like manipulation and locomotion where lack of robustness is a key limitation to these systems being deployed in the real world.


* denotes equal contribution. Representative papers are highlighted.

Generative Modeling of Residuals for Real-Time Risk-Sensitive Safety with Discrete-Time Control Barrier Functions

Ryan K. Cosner, Igor Sadalski, Jana K. Woo, Preston Culbertson, and Aaron D. Ames

IEEE International Conference on Robotics and Automation (ICRA), 2024. Under Review.

[PDF] [project page] [video]

By combining control barrier functions with a conditional variational autoencoder (CVAE), we develop a method for data-driven risk-sensitive control. Our method enables a quadrotor to keep a slung load above the ground despite having no explicit model of the load dynamics or measurements of the load state.

PONG: Probabilistic Object Normals for Grasping

Albert H. Li, Preston Culbertson, and Aaron D. Ames

IEEE International Conference on Robotics and Automation (ICRA), 2024. Under Review.

[PDF] [code]

We generalize our gradient-based grasp metric to the task of grasping objects with uncertain geometry. We show that our method can be used to generate robust grasps for objects with uncertain geometry in hardware, including objects modeled as NeRFs using an onboard camera.

Input-to-State Stability in Probability

Preston Culbertson, Ryan K. Cosner, Maegan Tucker, and Aaron D. Ames

IEEE Conference on Decision and Control (CDC), 2023. Accepted.


We generalize input-to-state stability for systems with unbounded disturbances (e.g., additive Gaussian noise). Using this (softer) stability notion we can formulate less conservative estimates of the tolerable disturbance, including for a 7-DOF walker with random step heights.

CATNIPS: Collision Avoidance Through Neural Implicit Probabilistic Scenes

Timothy Chen, Preston Culbertson, and Mac Schwager

IEEE Transactions on Robotics (T-RO), 2023. Under Review.


A Neural Radience Field (NeRF) density may be intepreted as a Poisson Point Process, essentially a continuous occupancy map. We use this theoretical insight to perform fast, chance-constrained path planning through NeRF scenes.

FRoGGeR: Fast Robust Grasp Generation via the Min-Weight Metric

Albert H. Li, Preston Culbertson , Joel W. Burdick, and Aaron D. Ames

IEEE/RSJ Conference on Intelligent Robots and Systems (IROS), 2023.

[PDF] [project page] [code]

We propose a differentiable, analytic grasp metric that enables collision-free grasp synthesis for dexterous hands in seconds (vs. minutes for existing methods). Can be used to post-process grasps generated from both simple heuristics and learning-based generative models.

Robust Safety under Stochastic Uncertainty with Discrete-Time Control Barrier Functions

Ryan K. Cosner, Preston Culbertson, Andrew J. Taylor, and Aaron D. Ames

Robotics: Science and Systems (RSS), 2023. Accepted.


Using tools from martingale theory, we provide bounds on the finite-horizion safety of a system whose controller imposes a control barrer condition in expectation.

Multi-Robot Assembly Scheduling for the Lunar Crater Radio Telescope on the Far-Side of the Moon

Preston Culbertson, Saptarshi Bandyopadhyay, Ashish Goel, Patrick McGarey, and Mac Schwager

IEEE Aerospace Conference, 2022.


Minimum-time multi-robot assembly planning for the Lunar Crater Radio Telescope via mixed-integer programming.

Vision-Only Robot Navigation in a Neural Radiance World

Michal Adamkiewicz*, Timothy Chen*, Adam Caccavale, Rachel Gardner, Preston Culbertson, Jeannette Bohg, and Mac Schwager

IEEE Robotics and Automation Letters (RA-L), 2021.

[PDF] [project page] [code] [video]

A Neural Radiance Function (NeRF), typically trained for novel view synthesis, can be used both for collision avoidance and robot localization with an onboard camera.

CoCo: Online Mixed-Integer Control via Supervised Learning

Abhishek Cauligi, Preston Culbertson, Edward Schmerling, Mac Schwager, Bartolomeo Stellato, and Marco Pavone

IEEE Robotics and Automation Letters (RA-L), 2021.


Using a classifier (supervised with offline data) to accelerate mixed-integer convex programming for robot planning. Yields 10-100x speedups over commerical solvers with similar solution quality.

Decentralized Adaptive Control for Collaborative Manipulation of Rigid Bodies

Preston Culbertson, Jean-Jacques Slotine, and Mac Schwager

IEEE Transactions on Robotics (T-RO), 2021.

[PDF] [code]

Even without knowledge of an object’s physical properties (including mass, friction, or geometry), a team of robots can use our proposed controller to track a desired 3D trajectory.

TrajectoTree: Trajectory Optimization Meets Tree Search for Planning Multi-contact Dexterous Manipulation

Claire Chen, Preston Culbertson, Marion Lepert, Mac Schwager, and Jeannette Bohg

IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2021.


Using search to find good contact sequences accelerates and improves the quality of contact-implicit trajectory optimizations for dextrous manipulation.

Learning Mixed-Integer Convex Optimization Strategies for Robot Planning and Control

Abhishek Cauligi*, Preston Culbertson*, Bartolomeo Stellato, Dmitris Bertsimas, Mac Schwager, and Marco Pavone

IEEE Conference on Decision and Control (CDC), 2020.

[PDF] [code]

We use a neural network classifier to yield real-time MICP solutions in robotics problems including obstacle avoidance, grasp selection, and through-contact planning.

Multi-Robot Assembly Planning via Discrete Optimization

Preston Culbertson, Saptarshi Bandyopadhyay, and Mac Schwager

IEEE/RSJ Conference on Intelligent Robots and Systems (IROS), 2019.

[PDF] [code] [video]

Multi-robot assembly planning can be posed as a mixed-integer program, including realistic physical modeling such as attachment directions and actuation limits for robots.

Decentralized Adaptive Control for Collaborative Manipulation

Preston Culbertson and Mac Schwager

IEEE International Conference on Robotics and Automation (ICRA), 2018. Best Manipulation Paper; Finalist, Best Multi-Robot Systems Paper.


By adapting their control gains, a team of robots can move an unknown payload to track a desired trajectory without explicit communication between agents.

Simultaneous Active Parameter Estimation and Control Using Sampling-Based Bayesian Reinforcement Learning

Patrick Slade, Preston Culbertson, Zachary Sunberg, and Mykel Kochenderfer

IEEE/RSJ International Conference on Intelligent Robots and Systems, 2017.


Posing control of an uncertain system as a belief-space MDP results in policies that can trade off exploration (actions that identify the system) and exploitation (actions that minimize cost).

Broader Impacts and Social Justice

I'm excited to explore the potential social impacts of robots and their application in society. I think it's crucial that roboticists anticipate negative social outcomes of our work and advocate for a more equitable and just society. If you have thoughts about social justice + robotics/AI, or are organizing research/events around this topic, feel free to email me (or DM me on Twitter).

Outside of robotics, my activism focuses mostly on issues around homelessness and housing, particularly in Los Angeles, where I live currently. I would love to connect with other roboticists / academics who work on these issues, so please reach out if you're interested.


My passion for engineering education began when I taught students on my solar car team skills ranging from machine design to CNC machining. As a grad student, I've had the chance to serve as a teaching assistant in graduate-level classes and mentor a diverse group of talented students. I hope to teach and design courses that blend engineering theory and practice and provide students with a broad, intuitive view of course concepts (including their potential social impacts).

Teaching Experience
  • Course Assistant, AA273: State Estimation and Filtering for Aerospace Systems, Stanford University, Spring 2018, 2021, 2022.
  • Course Assistant, AA273: Multi-robot Control, Communication, and Sensing, Stanford University, Winter 2018