From Ground to Muscle: The Start of a Learning Transition

Why change the question?

My engineering training comes from geotechnical engineering. The systems I used to study involved soil, structures, boundary conditions, and uncertainty. The questions drawing me now are about how the human body produces, controls, and adapts movement.

After completing my MSc, I worked from July 2024 to July 2025 as an Offshore Foundation Design Engineer in the Offshore Structural Technology Department of Goldwind’s R&D Center, supporting fixed offshore wind foundation work during pre-bid and bidding stages. Both research and engineering reinforced the importance I place on mechanical models, numerical methods, explicit assumptions, and validation.

The objects are different, but there is a clear connection: both require us to simplify a complex system into a model, state the assumptions, and test whether the model can explain the evidence.

That connection is the starting point of my move toward neuromusculoskeletal modeling. It is not a completed identity change; it is a learning process still underway.

What neuromusculoskeletal modeling means to me

Neuromusculoskeletal modeling (NMS) uses computational models to describe relationships among neural control, muscle force, skeletal motion, and the external environment.

It can support questions such as:

• estimating muscle forces and joint loads that are difficult to measure directly
• comparing how movement strategies affect tissues and joints
• informing rehabilitation, prosthetics, orthotics, and exoskeleton design
• connecting kinematics, dynamics, and motor control in one analytical framework

The applications are compelling, but the first priority for a learner is understanding the model boundaries: where the inputs come from, how parameters are chosen, and how sensitive a result is to its assumptions.

Where I am now

I am still building the foundation. The sequence is organized around capabilities needed for later research, not around collecting course certificates:

1. MIT 18.06 Linear Algebra — linear algebra and matrix methods
2. Andrew Ng Machine Learning — the basic framework of ML problems, models, and optimization
3. Karpathy Neural Networks: Zero to Hero — neural networks implemented from scratch
4. Mu Li’s Dive into Deep Learning (D2L) — systematic deep-learning models, code, and experiments
5. Shiyu Zhao’s Mathematical Foundations of Reinforcement Learning — entering RL through its mathematical foundations
6. Sutton & Barto, Reinforcement Learning: An Introduction — the classical RL framework
7. CS285 Deep Reinforcement Learning — my current stage

I do not want to present “currently learning” as “already mastered.” Project pages will state their progress, and notes will try to distinguish facts, interpretations, and plans.

Why make this site?

This site is less a showcase and more a public research notebook. I want it to do at least three things:

• record why a choice was made, not only the final output
• preserve assumptions, parameter sources, and failed paths for later review
• turn scattered learning into small projects that can keep moving

As my understanding changes, earlier notes may be revised. For someone crossing disciplines, leaving a visible trail of versions and corrections is part of the work.

Next step

My immediate priority is to complete CS285 and turn its key derivations, implementations, and experiments into reviewable notes. After CS285, I plan to use the MS-Human-700 series as my first integrated reproduction milestone; that work has not started yet.

This note is a starting point, not a conclusion.