I'm a Ph.D. student in Computer Engineering at Texas A&M University working on reinforcement learning for Physical AI: agents that have to perceive, predict and act in a physical world that is noisy, delayed and unforgiving.
My work covers the whole loop. I build the simulators and digital twins, the world models that predict what an action will do, the multi-agent policies that coordinate fleets, and the safety layers that keep them from crashing into each other. I apply it to drones, underwater vehicles, cars and robot arms. I got into this through underwater robotics, building AUV and ROV software for competition teams during my undergrad.
Away from the keyboard, I take fitness just as seriously. I train consistently, keep my diet dialled in, and I'm working towards six-pack abs. It's the same game as RL: a sparse, delayed reward that only shows up if you stay consistent.
💪 On your own health journey? Let's keep each other motivated. Add me on Google Health at zahidhussain909@gmail.com.
class Zahid:
role = "Ph.D. Computer Engineering @ Texas A&M University"
research = ["Reinforcement Learning", "Physical AI", "Multi-Agent RL", "World Models"]
platforms = ["🚁 drone swarms", "🌊 AUV fleets", "🚗 self-driving", "🦾 manipulation"]
toolbox = ["PyTorch", "MuJoCo", "PettingZoo", "V-JEPA 2", "Three.js"]
currently = "training one policy that can fly a drone team of any size"
off_duty = "training + clean diet, chasing a six-pack 💪"
fuel = "coffee ☕ > chai"
fun_fact = "I like to code and then stare at it in awe of its beauty."My research sits where reinforcement learning meets the physical world. I care about autonomous machines that have to coordinate, stay safe and make decisions from imperfect sensing, and I work on them from simulation through to policy.
| Multi-agent RL for aerial swarms. Policies that generalise across team size and team split instead of being locked to one configuration. Cooperative formation and coverage, competitive pursuit–evasion and capture-the-flag trained through self-play leagues, and RL for navigation that stays resilient when GPS can't be trusted. | Safe multi-agent RL for AUV fleets. Underwater, acoustic messages arrive late or not at all, so I work on control-barrier-function safety shields that keep a fleet collision-free under communication delay and packet loss, with RL learning the control on top. |
| World models for driving. Turning self-supervised video representations (V-JEPA) into action-conditioned world models that let a driving policy imagine the future before it acts, learned from real-world driving data. | Object-centric world models from pixels. Robot arms that see only camera frames but reason about objects, choose subgoals relative to them, and correct their own mistakes through interactive imitation (DAgger). |
| RL that decides where to measure. Constrained and risk-sensitive RL for placing a handful of sensors on physical fields you can't observe directly, so the field can be reconstructed accurately and robustly on a tight sensor budget. | Building the worlds agents learn in. Deterministic, vectorised simulators with realistic dynamics, wind, battery drain and lossy communication, plus MuJoCo environments and live 3D twins in the browser for watching policies act. |
I also rebuild research papers as interactive, scroll-driven explainers with redrawn diagrams, the maths kept intact, and mechanisms you can play with. Topics so far include PPO, REINFORCE, DreamerV3, I-JEPA and V-JEPA 2.1. They live at papers.zahid.win.
| RL & ML |
|
| Robotics & Systems | |
| Twins & Viz | |
| Infra & Writing | |
| Design |
- 🎓 Ph.D. Computer Engineering, Texas A&M University · B.Tech Electronics Engineering, ZHCET, Aligarh Muslim University (2019–2023)
- 🔬 Open to research collaborations and opportunities in RL, robotics and autonomous systems
- 📫 zahidhussain909@gmail.com
- 💪 Fitness accountability buddy? Add me on Google Health at the same email