Accelerating Multi-modal LLM Gaming Performance via Input Prediction and Mishit Correction

Ziyang Lin, Zixuan Sun, Sanhorn Chen +2

Real-time sequential control agents are often bottlenecked by inference latency. Even modest per-step planning delays can destabilize control and degrade overall performance. We pr…