●builderYou can use DDO to train agents that explore and maintain multiple successful paths for complex, multi-step tasks.
●researcherThis method addresses the limitations of trajectory-level outcome labels in preserving diverse successful branches during post-training.