I work on world models for robot manipulation — teaching policies to feel motion instead of just seeing frames. Currently building world-action models for dynamic bimanual tasks.
class Khalil:
focus = ["world-action models", "vision-language-action", "bimanual manipulation"]
stack = {
"core": ["Python", "PyTorch", "CUDA"],
"training": ["distributed training", "flash-attention", "distillation"],
"sim": ["RoboTwin", "SAPIEN", "CuRobo"],
"tools": ["uv", "Docker", "Linux"],
}
currently = "making robots act on how fast the world moves, not just where it is"Project page · Checkpoint
