alfayoung/robomme_1cuben_allcases
robomme_1cuben_allcases (VideoUnmaskSwap1CubeN) A single red cube is hidden under one of three cups; the cups are shuffled a variable number of times (Uniform{0..3} incl. 0-swap) and the robot must pick the cup now hiding the cube. Prompt is color-free: "watch the video carefully, then pick up the container hiding the cube". Monochrome. The cube is a fixed red in every episode (previously a per-episode random red/green/blue draw); the benchmark no longer varies cube color.… See the full description on the dataset page: https://huggingface.co/datasets/alfayoung/robomme_1cuben_allcases.
robomme1cubenallcases (VideoUnmaskSwap1CubeN)
A single red cube is hidden under one of three cups; the cups are shuffled a variable number of times (Uniform{0..3} incl. 0-swap) and the robot must pick the cup now hiding the cube. Prompt is color-free: "watch the video carefully, then pick up the container hiding the cube".
Monochrome. The cube is a fixed red in every episode (previously a per-episode random red/green/blue draw); the benchmark no longer varies cube color.
Sampling design. The 120 unique discrete cases (3 init positions × every swap-pair sequence of length 0..3) are each drawn 3 times from (object+cup init pose [seeded] × recovery mode {none, z, xy}), giving 360 episodes (112055 frames). Recovery-mode weights = [0.7, 0.15, 0.15] (~30% of episodes contain a deliberate failed grasp + recovery). Deep-verified: every episode reproduces its case.
Standard LeRobot v2.1 layout (panda, 10 fps). Precomputed VAE latents are not included.