Third Wave
Summer 2022 internship on autonomous forklifts: dynamics learning, tighter safety envelopes, and internal tooling to debug real test runs.
Autonomous forklifts
Tools
Python, PyTorch, TensorFlow, ROS, NumPy, OpenCV. Docker and Kubernetes for batch jobs; Git for everything else.
What I was doing there
At 16 I moved out to California for the summer and interned on the team building autonomous forklifts. Most days were models, logs, and simulator vs. real hardware, not slides. It was the first job where a sloppy assumption could mean a heavy machine doing something dumb in a warehouse, so we argued about safety a lot.
Actual work
Trained networks for 3-wheel Ackermann-style kinematics and chased down where sim and reality disagreed. Tightened safety envelopes so the vehicle didn’t flinch or overcorrect under load. Built a small internal tool: ingest test video, overlay planned vs. actual paths, skim a run without rerunning the whole stack.
What stuck
The useful part wasn’t hero ML. It was clear plots, boring metrics, and agreeing on what “safe enough” meant before we turned things loose.