VLA SOTA Leaderboard

Vision-Language-Action Model Benchmark Rankings

A comprehensive leaderboard tracking state-of-the-art Vision-Language-Action models on major robotics manipulation benchmarks.

📢 Latest Updates

Benchmarks

Include RL Models

LIBERO

0 Models

LIBERO is a benchmark for lifelong robot learning with 130 language-conditioned manipulation tasks across 4 task suites.

Primary Metric: Average Success Rate (%)

Top Performers

Meta-World

0 Models

Meta-World is a benchmark for meta-reinforcement learning and multi-task learning with 50 distinct robotic manipulation tasks.

Primary Metric: Average Success Rate (%)

Top Performers

CALVIN

0 Models

CALVIN is a benchmark for learning long-horizon language-conditioned tasks in a tabletop manipulation environment.

Primary Metric: Average Length (Avg. Len.)

Top Performers

LIBERO Plus

0 Models

LIBERO Plus is an extended benchmark testing robustness across 7 perturbation dimensions: Camera, Robot, Language, Light, Background, Noise, and Layout.

Primary Metric: Average Success Rate (%)

Top Performers

RoboChallenge

0 Models

RoboChallenge is a real-world robotic manipulation benchmark featuring challenging tasks in diverse environments.

Primary Metric: Score

Top Performers

RoboCasa-GR1-Tabletop

0 Models

RoboCasa-GR1-Tabletop is a benchmark built upon the RoboCasa simulation framework by GR00T-N1.5, tailored for GR-1 tabletop tasks, enabling the evaluation of generalist robotic policies in diverse household tasks.

Primary Metric: Average Success Rate (%)

Top Performers