CALVIN

University of Freiburg · 2022 · datasets.bot · datasets.bot page

One-liner. Simulated benchmark for long-horizon, language-conditioned Franka Panda manipulation: 4 envs, 34 tasks, 24h of play.

Setup

Schema

play sequences -> per-timestep .npy dicts -> {rgb_static, rgb_gripper, depth_static, depth_gripper, tactile, robot_obs, scene_obs, actions, rel_actions}; language annotations stored separately with task IDs and precomputed embeddings

Links