simulation and scientific computing.

robotics test bench

MuJoCo studies in feedback, coupled multibody dynamics, model-based control, and Jacobian mappings. public write-ups use a question, prediction, experiment, observation, and model-update structure.

C-1N // 03 · STRIDE

PPO learns joint corrections and gait cadence for a six-legged MuJoCo robot. walk_fast_500 reaches 0.888 m/s in the recorded mean-policy run, with no falls across 24 fixed flat-ground evaluations.

the recorded walking policy is separate from the historical browser stance controller. foot slip and contact fragmentation remain; terrain and push robustness are unproven.

tokenization, bigrams, and causal attention

the repository contains implementations for byte-pair tokenization, bigram language modeling, and causal attention in PyTorch. the linked tokenizer record includes Unicode round-trip assertions; the linked attention record shows shapes, causal masking, pre-normalization, and residual connections.

the numerical notebook records below are pinned to a reviewed 19222f1 revision. they document the study at that revision, not the full current scope of the repository.