Researchers propose Dynamic Mixture of Progressive Parameter-Efficient Expert Library (DMPEL) for lifelong robot learning.DMPEL utilizes a low-rank expert library and a lightweight router to combine experts dynamically for flexible behavior in lifelong adaptation.Coefficient replay is introduced to guide the router in accurately retrieving frozen experts, reducing catastrophic forgetting.Extensive experiments show that DMPEL outperforms existing methods in lifelong learning success rates with minimal trainable parameters and storage.