Preprint
Oct 2026
Gamma: Grounded Multi-Agent Memory with Mechanical Audit for Memory-Intensive Manipulation
Pretrained vision-language-action (VLA) policies fail on memory tasks, behaviour that depends on events observed long before the current frame, because their training data is overwhelmingly Markov and mapping episodic memory directly to action chunks is hard. Yet the same policy, finetuned once to follow a privileged o...
Yu-Dong Liu, Yueqian Lin, Zi-Jia Tang et al.
· 0 citations