Researcher, Sensetime
2 papers at NeurIPS 2025
We systematically investigate the design space and scaling property of native Multimodal Large Language Models and introduce a novel MLLM that achieves competitive performance against existing MLLMs.
This paper's HCRMP architecture implements an "LLM-Hinted" paradigm, using LLM hints for RL-based motion planning with crucial relative LLM-RL independence to improve driving safety and reliability.
