DriftQL is a one-step offline RL policy that replaces diffusion and flow denoising with a learned drift field: attraction keeps actions on the data, repulsion keeps them diverse, and the critic tilts them toward high value. State of the art on D4RL and OGBench, and robust to noisy datasets.
QWM conditions a single generative dynamics model on scale-invariant morphology features and trains locomotion policies entirely in imagination. Given the same morphology information, a model-free policy degrades on unseen robots while QWM transfers zero-shot; to our knowledge, it is the first world model to show zero-shot cross-embodiment transfer within the quadrupedal family.
VOCALoco predicts the viability and cost of transport of several pretrained locomotion skills from local heightmaps, then executes the safest, most efficient one. It improves robustness on stair ascent and descent over an end-to-end DRL policy and transfers to a real ANYmal-D quadruped.