CRIS-0 reportedly patches robot positioning functions after joint-limit errors

A post says CRIS-0 generates positioning functions from demonstrations using SAM3, RGB-D and inverse kinematics, then patches a function and retries when it hits a joint limit. The same poster describes a tablecloth task with rule-based pushing and positioning, learned grasping and folding, and checks that send a failed grasp back to repositioning. They say the learned policies come from CausalWM, a world model that predicts motion and 3D geometry before generating future frames.