Yang-Lun (Alan) Lai

yanglunlai@gmail.com

I'm an Engineer at Isuzu in California, where I work on motion planning and machine learning for autonomous driving.

I hold a master's degree in Electrical and Computer Engineering from the University of Michigan, where I was a member of the Image-Guided Medical Robotics Lab. There I conducted independent research on online trajectory generation for adaptive scanning, advised by Prof. Mark Draelos.

I earned my bachelor's degree in Biomechatronics Engineering from National Taiwan University (NTU). At NTU I worked with Prof. Ping-Lang Yen on agricultural robotics in the Robots and Medical Mechatronics Lab.

Outside of work I'm a singer with over five years of experience in a cappella and chorus, and I enjoy golf, scenic drives, and traveling.

Scholar  /  LinkedIn  /  Github

profile photo

Research

I'm interested in robot autonomy and machine learning — spanning autonomous driving, medical robotics, and agricultural robotics.

Autonomous Driving

Motion planning and machine learning for autonomous commercial vehicles.

Coming soon.

Adaptive scanning OCT method Adaptive scanning OCT results Figures from Rajani et al., Biomed. Opt. Express 16(1), 2025, © Optica Publishing Group, under the Optica Open Access Publishing Agreement.

Adaptive Scanning Optical Coherence Tomography

Predicting scene dynamics to drive where an OCT system scans next, for low-latency imaging of moving surgical scenes.

Dynamics-aware deep predictive adaptive scanning optical coherence tomography
D. M. Rajani, F. Seghizzi, Y.-L. Lai, K. G. Buchta, M. Draelos
Biomedical Optics Express, vol. 16, no. 1, p. 186, 2025
Journal article paper / full text

Adaptive scanning OCT with scene prediction and dynamics-aware scans
D. M. Rajani, F. Seghizzi, Y.-L. Lai, K. G. Buchta, M. Draelos
Optical Coherence Tomography and Coherence Domain Optical Methods in Biomedicine XXVIII, 2024, PC128302O
Conference abstract abstract

Low-latency surgical scene imaging with scene prediction and dynamics-aware scans
D. M. Rajani, F. Seghizzi, Y.-L. Lai, K. G. Buchta, M. Draelos
Investigative Ophthalmology & Visual Science, vol. 65, no. 7, pp. 5920–5920, 2024
Conference abstract abstract

Human-robot collaborative tea harvesting vehicle

Agricultural Robotics and Human-Robot Collaboration

Collaborative field machinery with adjustable autonomy, including Taiwan's first human-robot collaborative tea harvesting vehicle.
news coverage (CTS, in Chinese)

Coordinated Mechanical Operations in Fields
P.-L. Yen, Y.-L. Lai
Encyclopedia of Smart Agriculture Technologies, Springer, 2023, pp. 1–8
Book chapter chapter

A collaborative robot for tea harvesting with adjustable autonomy
Y.-L. Lai, P.-L. Chen, T.-C. Su, W.-Y. Hwang, S.-F. Chen, P.-L. Yen
Cybernetics and Systems, vol. 53, no. 1, pp. 4–22, 2022
Journal article paper

A human-robot cooperative vehicle for tea plucking
Y.-L. Lai, P.-L. Chen, P.-L. Yen
IEEE 7th International Conference on Control, Decision and Information Technologies (CoDIT), 2020
Conference paper paper

Supporting Vehicle System
P.-L. Yen, H.-Y. Hsu, Y.-L. Lai, P.-L. Chen, H.-Y. Chan
Taiwan Patent I737348, filed June 2020, issued December 2021
Patent

Projects
Robot arm trajectory optimization around two obstacles

Machine Learning for Dynamics Modeling and Trajectory Optimization of the Robot Arm

Learning a residual dynamics model of a robot arm and using it for trajectory optimization through an obstacle-cluttered workspace.

Machine Learning for Robotics

Kinodynamic RRT growing a search tree around obstacles to reach the goal

Kinodynamic RRT Motion Planning for a Planar Hovercraft Robot

Sampling-based motion planning that respects the dynamic constraints of a planar hovercraft.

Robotics
video / code

Loosely coupled pipeline: ORB-SLAM3 on images and IMU, wheel encoder and IMU through a delayed-state EKF, fused by GTSAM graph optimization Estimated trajectories compared against ground truth after graph optimization

Robot Localization using ORB-SLAM3 and Graph-Based Sensor Fusion

Loosely coupling ORB-SLAM3 trajectory estimates with wheel-encoder and IMU odometry from a delayed-state EKF, then refining the result by pose-graph optimization in GTSAM. On the NCLT dataset this reduced total trajectory RMSE from 61.73 m to 1.06 m.

Robotics
video / code

SR-aided NeRF method: a frozen FSRCNN produces super-resolved supervision targets for a trainable high-res NeRF NeRF novel-view render orbiting the ship scene

SR-aided NeRF: Super-Resolution Novel View Synthesis with Neural Networks

Generated high-resolution Neural Radiance Fields (NeRF) model from only low-resolution inputs by tightly integrated the Super-Resolution Convolutional Neural Network (SR-CNN) in the training pipeline. Reached 99% of the peak signal-to-noise ratio (PSNR) value achieved by NeRF models trained with high-resolution inputs.

Deep Learning for Computer Vision


Last updated August 2026. Website design based on Jon Barron's template.