Writing
Explanations, tutorials, build notes, and essays grounded in work that can be examined. Explore a concept, try an experiment, and follow the reasoning behind it.
Start with the interactive foundationsShowing 1–12 of 151 posts
Follow behavior cloning covariate shift through a frozen policy and a two-state motion model. Compare held-out expert actions with learner-state errors, inspect visited states, and test a policy class that can represent the expert.
Learn an inverse model from the learner’s own actions, infer missing expert labels, and test a cloned policy. See how poor exploration and sensing errors change the result.
Find which states an actuator can reach and which a sensor can reveal. Use exact cart motion, two held commands, matrix rank, and indistinguishable initial states.
Run DAgger imitation learning one round at a time. Collect expert labels on learner rollouts, retrain on the growing dataset, and compare equal label budgets.
Train two policies with the same labels and example count, then randomize one nuisance feature. Inspect the learned coefficients and compare frozen familiar and shifted tests to see what domain randomization changes.
A palletizing cell picks each case off a conveyor and builds a stable pallet. Learn its parts, the motion and sensing behind each placement, where cells fail, and who makes the hardware.
Compare imitation-learning methods by their data, learned quantities, and need for expert or environment access. Locate behavior cloning, DAgger, inverse RL, and GAIL.
Add an integral state to a cart controller, calculate three feedback gains, and compare constant-load tracking with reference feedforward. See why stable dynamics, accurate sensing, and sufficient force still matter.
Calculate LQR feedback for a cart, connect Q and R to position, velocity, and force, and separate accumulated cost from exact future cost.
Control a cart from noisy position readings. Connect Kalman filtering to LQR, compare matched disturbances, and test what changes when the model is wrong.
Build a Luenberger observer for a cart with a position sensor. Calculate observer gains, follow estimation error, and see why a small position residual can hide a velocity error.
Build a Lyapunov function, check its derivative, and prove convergence. See why a failed candidate can coexist with a stable system.