Yichen (James) Zhao is a third year PhD student studying Robotics and Mathematics in the Georgia Institute of Technology. His research interests include perception and motion planning, dynamical systems, distributed optimization, non-convex optimization, Riemannian optimization, Morse theory, sheaf theory, and algebraic topology.
We introduce a distributed generalized momentum algorithm that allows for arbitrary delays in communications and computations, and it converges faster than comparable existing algorithms.
Recommended citation: Ellie Pond*, Yichen Zhao*, and Matthew Hale (2025). "A Distributed Asynchronous Generalized Momentum Algorithm Without Delay Bounds." arXiv preprint arXiv:2508.08218. *Equal contribution. Download Paper | Download Bibtex
We introduce a framework for using time-invariant CBFs to form a time-varying CBF that guarantees safety while accelerating multi-robot navigation tasks. We first show that a time-varying convex combination of time-invariant CBFs forms a valid time-varying CBF, which renders the time-dependent safe set forward invariant.
Recommended citation: In preparation, authors: Yichen Zhao and Ellie Pond and Sean Wilson and Matthew Hale.
Published in Nonlinear Science and Control Engineering, Vol. 1, 2025
We investigate the accuracy, efficiency, and robustness of three recurrent neural network architectures: (i) next-generation reservoir computing, (ii) reservoir computing, (iii) long short-term memory (LSTM) for chaotic system prediction.
Recommended citation: Huaiyuan Rao, Yichen Zhao, Hsuan-Pin Chen (2025). "Predicting Chaotic System Behavior Using Machine Learning Techniques." Nonlinear Science and Control Engineering vol. 1, num. 1. Download Paper | Download Bibtex
We introduce an asynchronous nonlinear sheaf diffusion algorithm. Specifically, we show that under mild assumptions on the coordination sheaf and bounded delays in communication and computation, nonlinear sheaf diffusion converges to a minimizer of the Dirichlet energy of the coordination sheaf at a linear rate proportional to the delay bound.
Recommended citation: Yichen Zhao*, Tyler Hanks*, Hans Riess*, Samuel Cohen, Matthew Hale, and James Fairbanks (2025). "Asynchronous Nonlinear Sheaf Diffusion for Multi-Agent Coordination." arXiv preprint arXiv:2510.00270. *Equal contribution. Download Paper | Download Bibtex
In this paper, we propose STAR-Filter, a lightweight framework that employs starshaped set construction as a fast filter for convex region generation in collision-free space.
Recommended citation: Yuwei Wu, Yichen Zhao, Dexter Ong, and Vijay Kumar (2026). "Star-filter: Efficient convex free-space approximation via starshaped set filtering in noisy environments."arXiv preprint arXiv:2604.26626. Download Paper | Download Bibtex
Control barrier functions (CBFs) are widely used in multi-robot systems to augment control commands to enforce system safety. However, this safety guarantee can come at the cost of slower robot navigation. Therefore, in this work we introduce a framework for using time-invariant CBFs to form a time-varying CBF that guarantees safety while accelerating multi-robot navigation tasks. We first show that a time-varying convex combination of time-invariant CBFs forms a valid time-varying CBF, which renders the time-dependent safe set forward invariant. We also prove that the safe set is uniformly asymptotically stable. Then we introduce an algorithm that alters CBF shapes in real time to accelerate navigation tasks. The performance of the proposed algorithm is validated in hardware experiments in congested multi-robot environments, and it achieves up to a 43% reduction in the time required to complete navigation tasks compared to a baseline time-invariant CBF implementation.
Multi-agent systems are collections of interacting autonomous agents that are becoming increasingly prevalent with the rising popularity of learning-enabled autonomy. Achieving coordinated behavior in collaborative multi-agent settings raises fundamental challenges, including limited or structured communication, complex heterogeneous system architectures, and stringent safety requirements. In this talk, we develop theoretical tools and algorithms that address these challenges across communication, optimization, and safety. First, we leverage applied sheaf theory to model heterogeneous agents and tasks over complex communication topologies, and we propose an asynchronous nonlinear sheaf diffusion method for coordination. Second, we introduce a fully asynchronous momentum-based algorithm that accelerates convergence in parallel optimization without requiring known bounds on communication or computation delays. Finally, we develop a control barrier function (CBF) selection framework with time-varying CBFs called Delta-CBF that preserves safety guarantees while accelerating task completion.
