Xusheng Luo

I'm currently a Posdoctoral Fellow at Intelligent Control Lab of Carnegie Mellon University, working with Dr. Changliu Liu, starting from April 2023. I received the Ph.D. degree in Mechanical Engineering and Materials Science from Duke University in December 2020, under the supervision of Dr. Michael M. Zavlanos. Prior to it, I received the B.Eng. and M.S.E. degrees in Aerospace Engineering from the Harbin Institute of Technology, China, in 2015 and 2017, respectively.

I’m seeking a research position in academia or industry. Feel free to reach out if you think I could be a good fit!

Email  /  Scholar  /  Github

News

• 2025/05 - Serving as Session Chair of ICRA 2025
• 2025/04 - One paper accepted to RSS 2025
• 2025/04 - One paper accepted to Transaction on Cyber-Physical Systems (T-CPS)
• 2025/04 - One paper accepted to CAV 2025
• 2024/05 - One paper accepted to RA-L 2024
• 2024/04 - Selected as NSF Rising Star in Cyber-Physical Systems
• 2023/10 - One paper accepted to CoRL Workshop on Learning Effective Abstractions for Planning (LEAP)
• 2023/09 - One paper accepted to IROS Workshop on Formal methods techniques in robotics systems: Design and control
• 2023/04 - Joined Intelligent Control Lab (ICL) as a Postdoctoral Fellow

Research

I am a robotics and AI researcher specializing in developing intelligent and reliable autonomous systems. My work combines symbolic AI for high-level task planning with optimization-based motion planning, integrating both model-based and data-driven techniques. My research focuses on three fundamental questions:

1. Task Specification — How can we effectively specify tasks for autonomous systems?
2. Control Synthesis — How can we ensure robots behave as intended?
3. Verification/Certification — How do we formally verify that an algorithm meets its specified requirements?

This work addresses the TAMP problem in multi-robot settings where tasks are specified using expressive hierarchical temporal logic and task assignments are not pre-determined.

This is the first study to certify the robustness of large-scale, keypoint-based pose estimation given images in real-world scenarios.

This paper presents ModelVerification.jl (MV), the first comprehensive, cutting-edge toolbox that contains a suite of state-of-the-art methods for verifying different types of DNNs and safety specifications.

We proposed a neuro-symbolic paradigm of extracting task hierarchies from human instructions to facilitate multi-robot planning for complex, long-horizon tasks.

By leveraging the intrinsic structure of tasks, this paper introduced a hierarchical structure to LTL specifications. We employ a search-based approach to synthesize plans for a multi-robot system, accomplishing simultaneous task allocation and planning.

This paper proposes a a decomposition-based hierarchical framework for robotic takss under hierarchical Linear Temporal Logic (LTL) specifications.

This paper introduces a simulation-aided learning from demonstration (SaLfD) framework, enabling robots to learn and execute LEGO assembly and disassembly tasks from human demonstrations, successfully implemented and tested on a FANUC LR-mate 200id/7L robot.

This paper addresses the challenge of allocating tasks, defined by global Linear Temporal Logic (LTL) specifications, to diverse teams of robots.

This paper tackles the challenge of formal safety verification for stochastic cyber-physical systems (CPS) that use ReLU neural network controllers, aiming to identify initial states that ensure the system remains safe within a certain time frame.

This article introduces a new sampling-based linear temporal logic (LTL) planning algorithm that builds trees to explore the state-space, avoiding the need for complex discrete abstractions of robot mobility.

This paper proposes a new optimal graph-search algorithm designed for large-scale, linear time-invariant systems, which efficiently identifies malicious attacks and accurately estimates states.

This paper addresses the challenge of robot navigation in human-populated environments, aiming to create trajectories that are collision-free, dynamically feasible, and maximize human satisfaction by being responsive to human needs and avoiding discomfort.

This paper presents a novel framework for designing socially-aware robotic trajectories in human-populated environments, defining socially-awareness as the avoidance of human discomfort.

The paper addresses the single-agent indirect herding problem, focusing on a herder agent's task to guide a group of target agents to a specific goal.

This paper introduces an optimal control synthesis algorithm for Linear Temporal Logic (LTL) tasks, utilizing a novel approach that leverages experience from previous similar tasks.

Reviewer

Reviewer for T-RO, T-ASE, T-CNS, RA-L, L-CSS, RSS, ICRA, IROS, ACC, ICCPS, UR.