Zerui Li

Frontier models don't need retraining to master lifelong embodied tasks—structuring memory handovers and spatial consolidation across fresh reasoning sessions alone drives a 22.6-point surge in multi-task navigation success.

Agentic control in zero-shot navigation rivals industrial-scale policies, achieving up to 78% success with minimal interfaces and fewer resources.

Forget predefined maps: this VLN agent uses a learned, interactive world representation to achieve SOTA zero-shot performance and seamless sim-to-real transfer.