Tong Zhang

General-purpose VLMs can achieve state-of-the-art zero-shot robot manipulation by directly interpreting observations and issuing actions, eliminating the need for specialized policies or grounding tools.

Chasing large token-level teacher-student gaps in agent distillation is largely misguided: local divergence fails to predict task completion, whereas unpenalized minor deviations often trigger catastrophic multi-turn failure.

AgentSPEX transforms how we build and manage LLM-agent workflows, offering a modular and interpretable approach that outperforms traditional frameworks.