Part 1 · C++20
Learning to walk
Built bottom-up: matrices, a neural network, then world models, uncertainty from a group of 5 models, and planners. Proven first on a balancing cart-pole, then used to train a 12-motor biped in the MuJoCo physics simulator.
