Yicheng Rui

Tianyu Project (Full-stack Developer)

Jun 2023 - Present

Tianyu is a 1-meter robotic survey telescope in Lenghu, built to study exoplanets, variable stars, and transients.

• Develop observation-control, scheduling, and survey-analysis software.
• Built a distributed MySQL + RabbitMQ pipeline for image calibration, alignment, photometry, and light-curve production.
• Developed relative photometry approaching the theoretical noise limit on real telescope data, and estimated exoplanet and transient yields.

Project · Paper

Light-Curve Representation Learning with JEPA

Nov 2025 - Present

Irregular sampling and measurement noise make astronomical light curves challenging for general-purpose time-series models.

• Developed domain-informed multi-view self-distillation using raw light curves, periodograms, and phase-folded views.
• Learned transferable representations for classification, similarity search, parameter estimation, and photometric zero-point drift detection.
• Evaluated cross-domain adaptation to other irregular time-series datasets.

Preprint

Scientific AI Agents and Evaluation

Mar 2026 - Present

Variable-star vetting requires combining numerical tests, diagnostic plots, and source context into a traceable assessment.

• Developing a vision-language-model agent that combines Gaia, LAMOST, and VSX context with period searches, phase folding, and iterative visual diagnostics.
• Produces evidence-based reports with classifications, uncertainties, and follow-up priorities.
• Contributed two accepted Terminal-Bench Science v0.1 tasks: variable-star vetting and joint radial-velocity/astrometry inference; also served as a reviewer.

Code · Vetting task · Orbit task

LenghuSky-8: Cloud Segmentation and Weather Nowcasting

Oct 2025 - Feb 2026

Reliable cloud monitoring supports astronomical site characterization and autonomous telescope scheduling.

• Built an eight-year dataset of approximately 429,000 all-sky images from Lenghu, Qinghai.
• Performed absolute astrometric and altitude-azimuth calibration and constructed star-aware cloud masks.
• Benchmarked segmentation and nowcasting models, including DINOv3 linear probing and ConvLSTM; published in CVPR 2026 Findings.

Code and data · Paper

AstroSkyFlow: Astronomical Image Simulation

Nov 2024 - Jun 2026

Rare astronomical events and complex observing conditions make realistic simulations essential for validating survey pipelines.

• Developed a modular simulator combining variable celestial sources, atmospheric and instrument effects, and detector noise.
• Generates time-series sky images for machine-learning datasets and injection-recovery tests.
• Validated photometric and motion signals, including exoplanet transits and asteroid trails.

Paper

DeepAP: Fast Aperture Photometry

Nov 2024 - May 2025

Choosing suitable sources and apertures is a computational bottleneck in high-precision survey photometry.

• Co-developed a two-stage framework using a Vision Transformer to assess photometry feasibility and a ResNet to predict aperture size.
• Achieved approximately 59,000× speed-up over exhaustive aperture-size enumeration in the reported benchmark.
• Released the source code and trained model for reproducible photometry workflows.

Code · Paper

Mitigating Astrometric Bias

Dec 2021 - Jun 2023

Errors in stellar astrometry can bias barycentric corrections and obscure the weak radial-velocity signals of Earth-like planets.

• Developed a gradient-based method using PEXO to correct biases in stellar astrometric parameters.
• Studied the impact of astrometric systematics on ultra-precise radial-velocity measurements.

Paper

Robotics Competitions

2009 - 2019

Participated in RoboCup@Home and RoboCup Junior Maze, working on hardware control and autonomous navigation.

• World Champion, RoboCup Junior Maze (2018).
• National Champion, China Robot Competition and RoboCup@Home (2019).