Atul

Control · silicon · production software. Building real-time robot control, sim to hardware. ex-CTO @MoarMarket. Open to roles & collabs.

[Atul @SirGobta]

@bidhan @tanayj where does control systems fall into then? do you think its a solved problem in robotics?

[Atul @SirGobta]

7/ Writeup, source, and the four reasons it failed: https://t.co/e5kaJ5Gge9

[Atul @SirGobta]

1/ Three years in chip design, five years in software engineering - I’m moving into robotics now.
The video is a SO-101 5-DOF arm with an unknown weight in the gripper. It identifies the mass first, then tracks using it.

[Atul @SirGobta]

6/ So hardware identifies mass at rest from motor current, calibrates, freezes, then tracks.

90 g → 83 g. 180 g → 176 g.

Green is that estimate. Orange is uncalibrated — worse than none.

vs empty-model CT: arm RMS −31% / −45%.

[Mustafa. @oprydai]

how to get started in modern robotics:

don’t learn robotics like it’s 2015.

the stack has changed.
• math first → linear algebra, calculus, probability, optimization. robotics becomes much easier when the math stops looking mysterious.
• python + c++ → python for experimentation and ai, c++ for performance and real-time systems.
• ros2 → nodes, topics, services, actions, tf2, urdf. learn how an actual robot software stack fits together.
• simulation → mujoco, isaac sim, gazebo. build robots, destroy them, reset, repeat.
• control + estimation → pid, mpc, kalman filters, sensor fusion. intelligence is useless if the robot can’t control its own body.
• perception → cameras, lidar, radar, depth sensors. then learn cv, detection, tracking, slam.
• planning → a*, rrt, trajectory optimization, motion planning. perception tells you what exists, planning decides what to do.
• robot learning → imitation learning, reinforcement learning, vision-language-action models, world models.
• data → logging, replay, labeling, evaluation, synthetic data. modern robotics increasingly lives or dies by the quality of its data.
• hardware last → motors, encoders, actuators, batteries, embedded systems, mechanical design. now make everything survive reality.

then build one complete robot.

sensor → perception → state estimation → planning → control → actuator → feedback.

modern robotics isn’t one discipline.

it’s ai + software + math + control + hardware + physics, integrated into one machine that has to work in the real world.

[Atul @SirGobta]

@danieltao_eth DMed you

[Atul @SirGobta]

@edinsoncode DMed you

[Atul @SirGobta]

@PaulFrambot Gm! Sent you a dm

[Atul @SirGobta]

Having firsthand experience building lock-and-mint bridges, I fully understand the severity of this risk. A bug like this on one chain could easily have cascaded across the entire ecosystem through traditional mint/burn bridges.

This is exactly why we need more designs like Agglayer that isolate chain risk with pessimistic zk proofs.

[Mudit Gupta @Mudit__Gupta]

An arbitrary state write bug in Aptos chain was disclosed by @hexens today.

This is the worst kind of bug possible on a chain.

Why? Not only everything on the affected chain can be stolen but also most assets across all chain can be stolen.

Your stablecoins, LSTs, everything

[Atul @SirGobta]

@defyneric @FlexSuperApp Here is another ex-CTOs portfolio

https://t.co/g5GLWyXw8w

[Atul @SirGobta]

@Matthew_Graham_ @merkl_xyz @monad Got it. Thanks :)

[Atul @SirGobta]

@0xCster @0xTimepunk Let me know if this interests you.

https://t.co/sdeQk4VlrB

[Atul @SirGobta]

@orynth Built @MoarMarket, @HubbleExchange along various other side projects.
More info here : https://t.co/g5GLWyXw8w

[Atul @SirGobta]

@SuhailKakar @Polymarket I have 5+ year experience in deFi, built a perps exchange before. Open for integration or FDE role. DMed you

[Atul @SirGobta]

@Daya_HQ @alephile @0xPeejay Congrats team