Our research focuses on deformable agents, with training based on joint optimization, consisting of two stages: Skeleton Transform and Execution.
We propose a novel framework, Ske-Ex, which incorporates morphological and control knowledge in lifelong learning. This morphological knowledge includes the significance of morphology, such as limb size and length.
Our goal is to enhance the deformable agent's lifelong ability to adapt to sequential terrain tasks.
We now present the forgetting curves and generalization performance for order A, based on our benchmark. Order A : Gill → Land → Step
