Tag
A paper introduces TC-LeWM, applying regularization losses like SIGReg and VISReg to temporal latent residuals to enable multi-task learning in LeWM, significantly improving success rates on the LIBERO robot arm benchmark.