Selected research projects:
a. Efficient Optimisation Algorithms for Optimisation-based Control under Limited On-board Computation and Communication (common challenges in robotics)
We advance the algorithmic foundations and deployability of model predictive control (MPC, an optimisation-based method) for robotic systems. Methodologically, we developed scalable optimisation techniques tailored to real-time MPC control on resource-constrained hardware, with provable convergence and complexity guarantees. In parallel, we designed communication-aware distributed optimisation that remains reliable under realistic network conditions (limited bandwidth, quantisation, asynchrony, packet loss). We have demonstrated these methods on embedded and networked platforms, including aircraft and underwater robots, enabling faster and more robust MPC in practice. Relevant research papers can be found here.
Quantization design for distributed optimization. Y Pu, MN Zeilinger, CN Jones. IEEE Transactions on Automatic Control 2017
Complexity certification of the fast alternating minimization algorithm for linear MPC. Y Pu, MN Zeilinger, CN Jones. IEEE Transactions on Automatic Control 2016
Real-time distributed model predictive control with limited communication data rates. Y Yang, Y Wang, C Manzie, Y Pu. IEEE Transactions on Automatic Control 2025
Sub-optimal MPC with dynamic constraint tightening. Y Yang, Y Wang, C Manzie, Y Pu. IEEE Control Systems Letters 7 2024
b. Learning-based control for dynamical systems with non-asymptotic guarantee
We developed design-and-analysis frameworks for learning-based control of stochastic dynamical systems under uncertainty, with rigorous theoretical guarantees. The framework delivers: (i) a stabilising, certainty-equivalence adaptive control strategy for constrained non-linear systems; and (ii) non-asymptotic, high-probability estimation and stability guarantees by coupling stochastic control with statistical machine learning. Theoretically, we relax a common assumption in the literature — the existence of an initial stabilising controller. Practically, the stochastic adaptive-control framework accommodates diverse control laws, for example, deadbeat control and MPC. Relevant research papers include:
A Framework for Adaptive Stabilisation of Nonlinear Stochastic Systems. S Siriya, J Zhu, D Nešić, Y Pu. arXiv preprint arXiv:2511.17436, 2025
Non-Asymptotic Bounds for Closed-Loop Identification of Unstable Nonlinear Stochastic Systems. S Siriya, J Zhu, D Nešić, Y Pu. arXiv preprint arXiv:2412.04157, 2025
Learning-based adaptive control for stochastic linear systems with input constraints. S Siriya, J Zhu, D Nešić, Y Pu. IEEE Control Systems Letters 7, 1273-1278, 2022
c. Vision-based Perception and Navigation for Autonomous Underwater Vehicles (AUVs):
We advance vision-based perception and navigation for AUVs through a representation-learning lens, unifying perception and end–to-end navigation using sparse and dense representations. We first transfer robust in-air visual features to underwater settings via cross-modal distillation on physics-based synthetic imagery, strengthening sparse-representation-based perception for localization and state estimation. We then develop a physics-informed learning framework for dense-representation-based perception, adapting depth predictors with underwater imaging priors (attenuation, backscatter) to recover reliable scene geometry under turbidity and limited labels. Building on these learned representations, we design a deep-learning-based, image-only navigation policy that exploits depth-aware features to reach goals, maintain safe altitude, and avoid obstacles, without reliance on prebuilt maps. Relevant research papers include:
Physics-Informed Knowledge Transfer for Underwater Monocular Depth Estimation. J Yang, M Gong, Y Pu. European Conference on Computer Vision, 449-465, 2023
Duvin: Diffusion-based underwater visual navigation via knowledge-transferred depth features. J Yang, MQ Le, M Gong, Y Pu. arXiv preprint arXiv:2509.02983, 2025
Knowledge Distillation for Underwater Feature Extraction and Matching via GAN-synthesized Images. J Yang, M Gong, Y Pu. IEEE Robotics and Automation Letters 2025
Experimental videos can be found here and here.
d. Safety-guaranteed motion planning and control for robotic systems under complex uncertainties
We developed a unified approach to real-time, safety-guaranteed motion planning and control in the presence of model mismatch and complex environmental uncertainties. Our FaSTrack framework plans quickly on a simplified model while a higher-fidelity tracking model follows within a precomputed Hamilton–Jacobi tracking-error bound, providing formal guarantees on planning and tracking performance and safety without prior map knowledge and across diverse planner/model pairs. Relevant research papers include:
Towards Fast and Safety-Guaranteed Trajectory Planning and Tracking for Time-Varying Systems. S Siriya, M Chen, Y Pu. IEEE Transactions on Automatic Control, 2025
Fastrack: a modular framework for real-time motion planning and guaranteed safe tracking. M Chen, SL Herbert, H Hu, Y Pu, JF Fisac, S Bansal, SJ Han, CJ Tomlin. IEEE Transactions on Automatic Control 2021
Safety-guaranteed real-time trajectory planning for underwater vehicles in plane-progressive waves. S Siriya, M Bui, A Shriraman, M Chen, Y Pu. 2020 59th IEEE Conference on Decision and Control (CDC), 2020