Why robotics belongs on the trajectory.
The lab's long arc is intelligence plus systems. A robot is the place where intelligence meets the physical world. Once the AI work matures, the natural next surface is a body for the brain to live in. That is the throughline.
The other reason is range. Software-only engineers have a blind spot for the messiness of the physical world. The constraints are sharper, the feedback loops are slower, the failure modes are louder. Building inside those constraints, even at small scale, changes how you build everything else.
What I am doing about it today.
Reading the foundations: control theory, kinematics, the basics of embedded systems. Working in simulation before working with hardware, because simulation is the cheapest place to make mistakes. Watching what the open-source robotics community is shipping and learning from the patterns that hold up.
I am not pretending to have shipped a robot. I have not. The page exists to make the direction clear and the timeline honest.
The hardware plan.
Start small. A microcontroller, a few sensors, a single actuator. Build something that does one thing well, end to end. Sensor read, decision logic, motor output, feedback loop. The first build will be modest on purpose. The point is to learn the loop, not to impress anyone.
Scale from there. A wheeled platform with vision. A manipulator that picks up a known object. A small autonomous task that combines all of the above. Each step is a real build, not a demo.
A robot that does one thing reliably beats a robot that does ten things sometimes.
Where AI fits in.
The lab's AI work and robotics work are joined at the hip. Vision systems for object recognition. Local model inference for fast decision making on edge hardware. Agentic planning patterns for multi-step physical tasks. Every robotics experiment will pull from the AI experiments folder.
The reverse is also true. The constraints of physical hardware will sharpen the AI work in ways that pure software cannot. Latency is real. Failure is real. Tolerances are real. That pressure makes the system stronger.
What this folder will look like in a year.
Specific builds with specific learnings. A repository for each. Video of the thing working and video of the thing failing, because both teach the reader something. Notes on what the simulation got right and what only emerged from contact with reality.
The bar for this folder will be the same as the rest of the lab. Real, shipped, honest. Not promises.