Browse categories
Explore
Fiverr Pro
English
$
USD
M M Umair
@musabumair611
Aero Engineering Student, 4 years in ML, DRL based Control for UAV
Bangladesh
English, Bengali, Urdu, Hindi
About me
Aeronautical Engineering student at NUAA, China, with 4+ years of hands-on Python and AI experience. I build reinforcement learning agents (PyTorch, Stable Baselines3, RLlib), MATLAB simulations for engineering problems, and custom AI pipelines — trained and deployed on multi-GPU clusters. My current research: RL-based control for drag-free satellites (LISA-class missions), combining multibody dynamics, LQR optimal control, and deep RL. I deliver clean code, clear communication, and on-time results. Tell me your problem — I'll turn it into a working solution.
Skills
M M Umair
Offline •
Average response time: 1 hour
See my services
Data Labeling & Annotation
I will prepare clean labeled datasets for your ai or machine learning project
AI Implementation & Deployment
I will train a custom lora model for stable diffusion sdxl or flux on your images
Portfolio
Work experience
RL-Based Drag-Free Attitude Control for LISA
Nanjing University of Aeronautics and Astronautics • Part-time
Jul 2026 - Present • 3 mos
Ongoing research thesis — Nanjing University of Aeronautics and Astronautics (NUAA), started August 2026. Space-based gravitational wave observatories (LISA, Taiji) detect ripples in spacetime by holding free-floating test masses in near-perfect free-fall, with relative displacement kept inside a 2 mm electrode gap against solar radiation pressure and thermal disturbances that drift over time. Fixed-gain controllers degrade as the environment drifts — the exact problem this project tackles. My work, end to end: • Deriving the 17-state nonlinear multibody dynamics of a drag-free spacecraft (spacecraft attitude, two 6-DOF test masses, two optical assemblies) using Kane's method — symbolic derivation in Python/SymPy, simulation in MATLAB/Simulink, validated against published reference models. • Designing the baseline optimal controller (LQR) via Jacobian linearization. • Building a causal time-frequency disturbance analysis pipeline (STFT/wavelet regime features). • Training a Soft Actor-Critic reinforcement learning agent (PyTorch, Stable Baselines3) that augments the baseline by scheduling its gains online across physics-derived linearized models — adapting to the drifting environment while preserving the baseline's stability guarantees. Training is fully offline, in simulation. • Monte Carlo evaluation over the full LISA noise model set to quantify performance against the fixed baseline. Tools: Python (SymPy, PyTorch, Stable Baselines3), MATLAB/Simulink, multibody dynamics, optimal control, deep RL.
