Search papers, labs, and topics across Lattice.
CogAI Lab, Moscow, Russia
3
0
6
Isolating recurrence in VLA models reveals it can double success rates in partially observable tasks without complicating architecture or introducing auxiliary losses.
Forget meticulously annotating subtasks – SuperIgor lets language models self-learn to generate and refine instruction-following plans through RL feedback.
A single transformer can master StarCraft, Football, and POGEMA, suggesting we can unify MARL under one foundation model.
