ELMUR: External Layer Memory with Update/Rewrite for Long-Horizon RL Problems

Researchers developed ELMUR, a new AI architecture that uses external memory to help robots make better decisions over extremely long time periods. The system achieved 100% success on tasks requiring memory of up to one million steps and nearly doubled performance on robotic manipulation tasks compared to existing methods.

• →ELMUR extends effective decision-making horizons up to 100,000 times beyond standard attention windows
• →The system achieved 100% success rate on synthetic T-Maze tasks with corridors up to one million steps
• →On robotic manipulation tasks, ELMUR nearly doubled baseline performance and achieved best results on 21 out of 23 tasks
• →The architecture uses structured external memory with bidirectional cross-attention and LRU memory modules
• →ELMUR outperformed baselines on more than half of POPGym benchmark tasks