RoboVerse

RoboVerse is the dataset, benchmark, task-pack, and learning layer that sits on top of the MetaSim simulator framework. This section brings together the task inventory, asset descriptions, benchmark protocol, and the imitation- and reinforcement-learning workflows.

---

Quick Navigation

Tasks Overview

Browse the full task inventory across every benchmark namespace.

Tasks Overview

Robots, Objects, Scenes

Robot configurations, object assets, and scene definitions used by tasks.

Robots in RoboVerse

Imitation Learning

Diffusion Policy, ACT, OpenVLA, SmolVLA, RDT, Octo and how to add your own.

Diffusion Policy

Reinforcement Learning

PPO, FastTD3, SAC, TD3, SkillBlender, humanoid locomotion stacks.

PPO

Benchmark Protocol & Results

Standard evaluation protocols and the latest benchmark results.

Benchmark Overview

MetaSim docs ↗

Simulator core: concepts, handlers, API reference, troubleshooting.

/metasim/