When Validation Stops Learning: Auditing Update Admission for Continual Embodied Agents

Summary

This post summarizes an arXiv paper (2509.05819) by Qinzhen Ma and Ruihai Wu on update admission for continual embodied agents. Independent evaluation can reject harmful policy updates but may also block useful continual learning. The authors argue that update admission must be assessed through both error control and retained learning opportunities under a stated interaction budget. They identify a concrete failure mode: a range-based confidence gate cannot certify unchanged old-task behavior even with substantial budgets, while a standard paired-binomial construction reduces this burden when outcome disagreements are rare. They also specify certified historical-reference promotion and a round-level missed-opportunity metric. In a one-step pushing diagnostic with 32 seeds, fresh paired checks admitted 31.6% of a common update stream at 2,000 episodes per stage, versus zero for the range-based gate; unconditional replay still learned better in closed-loop runs. A learning-dynamics stress test separates model bias from feedback selection errors. The contribution is an admission audit protocol with analytical and synthetic evidence; physical robot and VLA validation remain open problems.