Search papers, labs, and topics across Lattice.
5
0
8
8
Agents struggle with long-horizon tasks, achieving only a 15.2% success rate even with advanced models, highlighting a critical gap in current AI capabilities.
Pretrained VLA models can be transformed into safer navigators, reducing near-collisions by over 86% through better alignment of their internal representations with action outputs.
LLMs can learn to play complex games far more effectively by co-evolving a skill bank with a decision-making agent, enabling consistent long-horizon decision-making.
Stop bloating your agent's context window: a dependency-aware skill graph slashes token costs by 37% and boosts reward by 43% compared to naive skill loading.
VLMs can now self-evolve from *zero* data, thanks to a multi-agent RL framework that synthesizes its own visual concepts and reasoning tasks.
