Trajectory-Level Automatic Curriculum Learning for Legged Locomotion on Unstructured Terrain

A new arXiv preprint (2608.16164) introduces Trajectory-Level Automatic Curriculum Learning (TACL), a framework for training legged robots to navigate unstructured terrain. The method addresses the challenge of curriculum design when terrain lacks explicit difficulty ordering. TACL generates training tasks directly from terrain maps, with an evaluator learning a difficulty function for the current policy and a sampler proposing new trajectories guided by the evaluator. This closed-loop system iteratively matches the curriculum to the policy's capabilities, aiming to improve generalization over heuristic curricula that can overadapt to fixed perceptual patterns. The work is relevant to robotics and AI, though it is not directly related to contemporary art.