InteractVLA

Agent harness that turns a frozen flow-based VLA (Meituan-Robotics-0) into interactive tools an LLM planner calls at test time. Its reported results use recursive self-improvement: the agent practises over repeated attempts in practice environments before evaluation. Only the interaction code is released; it needs a separately supplied VLA checkpoint.

Renmin University of China & Meituan · Nov 17, 2026 · Agent system · built on Meituan-Robotics-0Report an error
2Benchmarks reported
Nov 17, 2026Released
APIWeights
Agent systemType

Benchmark results2

RoboCasa365Simulation
  • 89.4%, self-reported by the authors, not a leaderboard entryPaper / model card ↗Nov 2026· self-reported, recursive self-improvement· non-standard
89.4%, self-reported by the authors, not a leaderboard entrynot ranked
LIBERO-PlusSimulation
  • 99.4%, self-reported by the authors, not a leaderboard entryPaper / model card ↗Nov 2026· self-reported, recursive self-improvement· non-standard
99.4%, self-reported by the authors, not a leaderboard entrynot ranked
Not reported on 13 benchmarks
Wrong or missing number? Report an error