Gilbert Bahati
Gilbert Bahati
Embedded Files
I am a PhD candidate in Mechanical Engineering at the California Institute of Technology (Caltech) and part of the AMBER Lab, advised by Dr. Aaron D. Ames. My research is focused on perception-based safety and nonlinear control for robotic systems. I am particularly interested in leveraging tools from applied analysis and partial differential equations to address challenges in real-world autonomy that arise from perception-driven tasks, and validating these methods on hardware across various robotic platforms (e.g., humanoids, quadrupeds, drones etc).
Prior to coming to Caltech, I obtained my BS in Civil Engineering (Systems) at the University of California, Berkeley (UC Berkeley). While there, I was part of the Berkeley Artificial Intelligence Research (BAIR) Lab and worked with Dr. Alexandre M. Bayen on reinforcement learning (RL), optimal control and numerical methods applied to mixed-autonomy traffic, using conservation laws (PDEs) and FLOW - I was a member of the CIRCLES project.
CV / Linkedin / Google Scholar
Research
Research
Risk-Aware Safety Filters with Poisson Safety Functions and Laplace Guidance Fields
Gilbert Bahati, Ryan M. Bena, Meg Wilkinson, Pol Mestres, Ryan K. Cosner, Aaron D. Ames
Submitted (In review) IEEE American Control Conference (ACC), 2026.
Different obstacles demand different levels of caution. Given a semantic understanding of an environment with associated risk levels of environmental features, we propose a method for synthesizing perception-based, risk-aware safety filters that guarantee safety while prioritizing avoidance of higher-risk obstacles.
The proposed approach is composed of steps: encoding an understanding of the environment via Poisson’s equation, and associated risk via Laplace guidance fields. Specifically, we first solve a Dirichlet problem for Poisson’s equation to generate a safety function that encodes system safety as its 0-superlevel set. We then solve a Dirichlet problem for Laplace’s equation to construct a guidance field that encodes variable levels of caution around obstacles—by enforcing a tunable flux boundary condition.
Geometry-Aware Predictive Safety Filters on Humanoids: From Poisson Safety Functions to CBF Constrained MPC
Ryan M. Bena, Gilbert Bahati, Blake Werner, Ryan K. Cosner, Lizhi Yang, Aaron D. Ames
IEEE-RAS International Conference on Humanoid Robots (Humanoids), 2025. Best Oral Paper Finalist.
We propose a perception-based predictive safety filter: a nonlinear model predictive control (MPC) algorithm for online trajectory generation with geometry-aware safety constraints based on control barrier functions (CBFs).
We extend Poisson safety functions to incorporate temporal changes in the domain by reformulating the static Dirichlet problem for Poisson’s equation as a parameterized moving boundary value problem. Furthermore, we employ Minkowski set operations to lift the domain into a configuration space that accounts for robot geometry.
Dynamic Safety in Complex Environments: Synthesizing Safety Filters with Poisson’s Equation
Gilbert Bahati, Ryan M. Bena, Aaron D. Ames
Robotics: Science and Systems (RSS), 2025.
We present an algorithm for generating safe sets in real-time from perception data by leveraging elliptic partial differential equations (PDEs), specifically Poisson’s equation. We present Poisson Safety Functions—functional representations of the environment that characterize safety, by solving a boundary value problem for Poisson’s equation. We show their ability to incorporate desired gradient behaviors customized to different regions of a domain, and demonstrate their effectiveness on humanoid and quadrupedal robotic platforms.
Control Barrier Function Synthesis for Nonlinear Systems with Dual Relative Degree
Gilbert Bahati, Ryan K. Cosner, Max H. Cohen, Ryan M. Bena, Aaron D. Ames
64th IEEE Conference on Decision and Control (CDC), 2025
We provide a constructive framework for synthesizing CBFs for systems with dual relative degree—where different inputs influence the outputs at two different orders of differentiation; this is common in systems with orientation-based actuation, such as unicycles and quadrotors.
Sample-and-Hold Safety with Control Barrier Functions
Gilbert Bahati, Pio Ong, Aaron D. Ames
IEEE American Control Conference (ACC), 2024.
We examine the assumption that high sampling frequency leads to minor safety violations for controllers deployed on digital platforms (i.e., zero-order hold implementations). We propose an alternative approach to maintaining safety of such systems by modulating the sampled control input to ensure a more robust safety condition - avoiding any violations.
Characterizing Smooth Safety Filters via the Implicit Function Theorem
Max H. Cohen, Pio Ong, Gilbert Bahati, Aaron D. Ames
IEEE Control Systems Letters (L-CSS), 2023.
We present a general characterization of smooth safety filters – smooth controllers that guarantee safety in a minimally invasive fashion – based on the Implicit Function Theorem. This characterization leads to families of smooth universal formulas for safety-critical controllers that quantify the conservatism of the resulting safety filter.
Pio Ong, Gilbert Bahati, Aaron D. Ames
61th IEEE Conference on Decision and Control (CDC), 2022.
[pdf]
We generalize the event-triggered control concept to include state triggers where the controller can be turned off - intermittent control. Using certificate functions (Lyapunov or Barrier Functions), we show that our design of intermittent trigger laws guarantee stability or safety.
Work was done in collaboration with NASA-JPL.
Hamilton-Jacobi Reachability
Multi-Adversarial Safety Analysis for Autonomous Vehicles
Gilbert Bahati, Marsalis Gibson, Alexandre M. Bayen
Robotics: Science and Systems (RSS): Robust Autonomy workshop, 2020.
[pdf] / [video] / [HJ Reachability toolbox]
We study the reduction of conservativeness in Hamilton-Jacobi safety analysis by examining trade-offs between different modeling strategies.
We demonstrate how by introducing structure in the interactions between autonomous vehicles and surrounding vehicles in dense driving scenarios, we are able to uncover safe strategies.
Work was done under the CIRCLES project.
Page updated
Google Sites
Report abuse