LEARNING: motion
ABOUT
Most of my work starts with a question about what is happening underneath the abstraction.
The path was non-linear. The questions became harder. The engineering became increasingly serious.
Fahim Faisal builds machines that perceive the world and act inside it. He works across robotics, computer vision, embedded systems, control, and AI — usually starting from a simple question: what is actually happening underneath?
THE TIMELINE
Each chapter built on the previous one. Then the questions became harder.
-
-
How do machines move?
- LEARNING: perception
-
How does a machine see?
-
[03]
MACHINE VISION
LEARNING: estimation -
How do you know the map is true?
-
[04]
2025 SUBTEAM LEAD
LEARNING: integration -
How do subsystems agree?
-
[05]
ARCHITECTURE
LEARNING: leadership -
Where should intelligence live?
-
[06]
ROBOSUB 2026 / ENGINEERING LEAD
LEARNING: architecture -
How do I make the whole machine reliable?
-
[07]
MONGOL TORI
[VERIFY]LEARNING: autonomy -
What should I build if I control the architecture?
-
[08]
INDEPENDENT SYSTEMS / UE
[VERIFY]LEARNING: independence -
What happens when nobody is watching?
- [NEXT]
THE TWO TRAJECTORIES
There were periods when the formal path did not look clean. University transitions, imperfect grades, the sense of running late while other people's robots were already in the water. I was falling behind in one graph while moving forward in another.
The two paths did not move at the same speed. But the engineering trajectory was becoming clearer — more difficult problems, broader scope, heavier responsibility. The machine was being built at a different pace than the résumé, and the machine was what mattered.
THE DESCENT
Five layers, from intent down to the world. The descent is the default mode.
[LAYER 01] HIGH-LEVEL what the machine is trying to do
- mission intent — the goal the machine is given
- behavior selection — which mode applies right now
- planning — a route through the world, or through the problem
- perception goals — what the machine needs to see in order to act
[LAYER 02] MID-LEVEL the logic that decides, in real time
- control loops — the feedback that keeps the machine honest
- state estimation — where the machine thinks it is
- sensor fusion — several imperfect senses, one usable picture
- decision logic — what to do with the picture
[LAYER 03] LOW-LEVEL the code that talks to the metal
- drivers — the first translation layer
- interrupts — the hardware asking for attention
- real-time scheduling — the promise that the loop runs on time
- register-level code — where the abstraction ends
[LAYER 04] HARDWARE the parts that carry the current
- boards — the silicon and the traces between it
- sensors — the machine’s senses
- actuators — the machine’s muscles
- power — everything downstream of the battery
[LAYER 05] PHYSICAL the world that does not negotiate
- water — denser, darker, no GPS
- air — the medium the machine has to move through
- friction — the world pushing back
- latency — the time between sensing and acting
THE LOOP
- Build
- Observe
- Fail
- Understand
- Rebuild
Each project is one loop at a deeper level.
CONTRADICTIONS
Six pairs that are all true at once.
AMBITIOUS the machine should do more than it was asked
INSECURE the fear that it is not enough
SOPHISTICATED SYSTEMS estimation, control, perception working together
HATES UNNECESSARY COMPLEXITY the simplest thing that survives contact with reality
HARDWARE-LEVEL registers, interrupts, the physical layer
HIGH-LEVEL behavior, intent, the whole system
OPEN SOURCE knowledge should compound
PROPRIETARY-GRADE SERIOUSNESS the work is held to a production standard
TECHNICALLY SERIOUS the engineering is not a joke
ABSURD HUMOR the jokes are
FAILURE the detector worked, the vehicle didn’t
GROWING AMBITION each failure raised the ceiling
MANIFESTO
[M] I like machines that have to deal with reality.
[M] I like understanding what sits beneath the abstraction.
[M] I like taking expensive or opaque systems apart conceptually and rebuilding them from first principles.
[M] I like robotics because there is nowhere to hide: eventually the machine has to move.
[M] I like software because it lets an idea become behavior.
[M] I like hardware because physics gets the final vote.
[M] I like open source because knowledge should compound.
[M] I like difficult problems because they expose what I don’t understand.
[M] And I am still building.
BANGLADESH
Dhaka, Bangladesh. This is where the machine is being built. The tools are the same; the constraints are different. The work does not wait for a different address.
INFLUENCES
Not idols — sources. What I borrow is the approach: idea, implementation, working system.
GEORGE HOTZ the willingness to open the box and rebuild it — first principles over received wisdom
ANDREJ KARPATHY the discipline of teaching from the bottom up — mechanisms, not magic
COMMA.AI the calibration mindset — there is no magic, just long calibration
THE OPEN-SOURCE COMMUNITY the habit of publishing the work so knowledge compounds