High school student · Self-directed researcher · AI × Spacecraft Control
I am interested in spacecraft autonomy: how software, control, and learning systems can help a spacecraft make better propulsion decisions in uncertain environments.
My work focuses on building reproducible simulations, explicit controllers, and engineering logs. I treat failure as data, not as something to hide.
Trajectories drift. Controllers adapt. Failure is telemetry.
A long-term research-engineering project exploring autonomous orbital insertion and propulsion control in a physics-based 2D orbital environment.
Phase36C — Non-Crossing Geometry Diagnosis
The project evolved from simple rule-based controllers and PPO experiments into a structured investigation of orbital transfer geometry and recoverability.
Recent milestones:
- Phase34: Developed a robust post-cross synchronization controller and recoverability framework.
- Phase36B: Benchmarked four transfer-family designs across a reduced orbital test suite.
- Phase36C: Diagnosed the remaining non-crossing cases and identified the primary bottleneck as upstream crossing-generation, not post-cross stabilization.
- Multiple transfer families converge to the same crossing basin.
- Recoverability after crossing is largely solved in the current simulator.
- The remaining challenge is generating new Phase34-compatible crossings from difficult initial conditions.
- Geometric metrics can improve without necessarily creating new crossings.
- Physics-based orbital simulation environment
- Explicit phase-structured spacecraft controllers
- PPO and imitation-learning baselines
- Transfer-family benchmark framework
- Recoverability and crossing-basin analysis tools
- Failure-mode diagnostics and benchmark evaluation pipelines
- Detailed engineering and research logs documenting both successes and failures
Which parameterized transfer trajectory can generate new recoverable crossings among the remaining non-crossing orbital cases?
The long-term goal is to investigate how planning, control, and learning-based methods can cooperate in autonomous spacecraft guidance, eventually extending beyond reactive control toward trajectory-generation and decision-making systems.
🔗 Repo: https://github.com/Sean-ZhiXin-Li/spacecraft-ai-controller
A unified engineering workspace for building lab-ready systems habits through real execution.
Current focus:
- Python engineering from scratch
- Linux / WSL workflow
- Git discipline and repo hygiene
- Debugging and reproducibility
- Config → run → metrics → verification pipelines
- Small AI / simulation-oriented engineering experiments
This repository is not designed to look flashy. It is designed to show process: how I set up systems, run experiments, record outputs, debug failures, and gradually turn isolated scripts into reproducible engineering workflows.
It connects directly to my larger AI orbital-control project by strengthening the engineering foundation behind reliable experiments: environment setup, command-line workflow, metrics, validation, and documentation.
🔗 Repo: https://github.com/Sean-ZhiXin-Li/the-new
I am also building foundations in areas that support spacecraft autonomy:
- Control systems
- Robotics
- Embedded systems
- Reinforcement learning
- Simulation engineering
- Basic CAD / CubeSat structure exploration
🔗 Repo: https://github.com/Sean-ZhiXin-Li/tech-foundations
I care about process integrity.
- Logs preserve the reasoning path
- Bugs reveal structure
- Negative results are evidence
- Reproducibility matters more than appearance
- Progress is often spiral, not linear
Open-ended engineering projects taught me persistence more than drills or competitions did.
I want to keep working toward spacecraft autonomy: controllers that can adapt, degrade gracefully, and survive uncertainty.
My current work is still early and limited to simulation, but it gives me a concrete path to learn control, AI, physics, and engineering through one long-term system.
- Email: tlizxin209625@gmail.com
Focus: AI × Spacecraft Control · Simulation · Reproducible Engineering