OmniJARVIS: Unified Vision-Language-Action Tokenization Enables Open-World Instruction Following Agents Paper • 2407.00114 • Published Jun 27 • 12