Lei Zheng

Research Fellow · National University of Singapore

Safety-critical autonomy · robot safety · motion planning · safe learning-based control

About me Publications

🔥 News

Projects

Research on safe autonomy, planning, and learning

*
DNKD: Kinova Gen3 hardware and MuJoCo Walker2d locomotion

Real-Time Generative Robot Control via Learning-Based Policy Distillation

Distills multistep diffusion robot policies into a one-step Koopman student for millisecond-level closed-loop control. Validated on D4RL locomotion benchmarks and a physical Kinova Gen3 manipulator for obstacle-aware reconfiguration. Project page: fdkoopman.github.io.

Occlusion-Aware Contingency Safety-Critical Planning for Autonomous Vehicles

This research introduces a occlusion-aware contingency safety-critical planning approach for safety-critical autonomous vehicles. Leveraging reachability analysis for risk assessment, forward reachable sets of occluded phantom obstacles are computed. The occlusion-aware contingency planner then constructs multiple locally optimal trajectory branches (each tailored to different risk scenarios), and a shared consensus trunk is generated to ensure smooth transitions and motion consistency.

Safety-critical Agricultural Drone

To achieve high-speed autonomous flight of aerial vehicles and realize high-performance precision spraying in precision farming. Trajectories must be generated in real-time to avoid collision and be close to the reference spraying path. Because of the high navigation speed, short sensing range, and unknown environments, response time is extremely limited, making generating high-quality trajectories a significant challenge.

Safe and Real-Time Consistent Planning for Autonomous Vehicles in Partially Observed Environments via Parallel Consensus Optimization

This research introduces a consistent parallel trajectory optimization (CPTO) approach for real-time, consistent, and safe trajectory planning for autonomous driving in partially observed environments. The CPTO framework introduces a consensus safety barrier module, ensuring that each generated trajectory maintains a consistent and safe segment, even when faced with varying levels of obstacle detection accuracy. We validate our CPTO framework through extensive comparisons with state-of-the-art baselines across multiple driving tasks in partially observable environments. Our results demonstrate improved safety and consistency using both synthetic and real-world traffic datasets.

Barrier-Enhanced Homotopic Parallel Trajectory Optimization for Safety-Critical Autonomous Driving

Enforcing safety while preventing overly conservative behaviors is essential for autonomous vehicles to achieve high task performance. In this project, we propose a barrier-enhanced homotopic parallel trajectory optimization approach with over-relaxed alternating direction method of multipliers for real-time integrated decision-making and planning in cluttered driving environments. Through a series of experiments, the proposed development demonstrates improved task accuracy, stability, and consistency in various traffic scenarios using synthetic and real-world traffic datasets. We have also implemented our algorithms on an autonomous car, successfully achieving vehicle following, lane changing, overtaking, and cruise driving in dense traffic flow simulations based on ROS2.

High-Speed Agricultural Flight & Multi-Agent Precision Farming

Robust real-time trajectory generation for agricultural aerial vehicles to enable safe high-speed autonomous flight and precise spraying under short sensing range and unknown environments. Memory-efficient replanning supports smooth return and dynamic height adjustment, and extends to multi-agent safety-critical navigation for precision farming. Demo: YouTube.

Safe Learning

Designed efficient incremental Gaussian Processes accounting for airflow uncertainties. The wind disturbance caused by the external environment is estimated to improve flight safety and control stability in cluttered environments. Following that, the estimated wind disturbance is used to compensate for the associated control error.

Connected Cruise Control in Mixed Traffic Flow

For safety-critical vehicles in mixed traffic flow where most vehicles are human-driven, each autonomous vehicle keeps tracking its front vehicle at a desired constant speed while maintaining a safe following distance with it in normal situations. However, when a vehicle decelerates urgently in unexpected situations, the vehicle behind has to reduce its speed to avoid collision with the front vehicle. In these cases, there exists a conflict between safety and stable high-performance tracking. For safety-critical autonomous vehicles, safety must not be violated and the tracking errors should be kept as small as possible.

Experience

National University of Singapore 2026 - Now
Research Fellow @ NUS

Collaborating with Prof. Armin Lederer on generative models and learning-based safety filters, spanning real-time diffusion-policy distillation and task-adaptive robustness for humanoid robots in unstructured environments.

Carnegie Mellon University — Robotics Institute 2025
Visiting Scholar @ CMU, Robotics Institute

Collaborated with Prof. Changliu Liu on fundamental robot safety, spanning formal methods and learning-enabled safe control for autonomous systems.

The Hong Kong University of Science and Technology 2022 - 2025
Ph.D. in Robotics and Autonomous Systems @ HKUST

Worked with Prof. Jun Ma and Prof. Michael Yu Wang on safety-critical motion planning, spanning occlusion-aware contingency planning and trajectory optimization for autonomous vehicles.

XAG 2021 - 2022
Senior Robotics Engineer @ XAG

Led real-time trajectory replanning and safe backup policies for agricultural UAVs, spanning motion control, flight-data analysis, and technology transfer to field systems operating in 60+ countries.

Publications

Selected recent journal and conference papers

Contact

Collaborations, speaking invitations, and opportunities

  • 4 Engineering Drive 3, National University of Singapore, Singapore 117583
Visitor map
Locations of visitors to this page