DAHLIA is a novel framework that enables robots to perform long sequences of tasks using natural language commands without requiring domain-specific training data.

• Uses vision-language models to provide continuous feedback during task execution
• Implements closed-loop control that adapts to changing environments and recovers from failures
• Achieves 95% success rate on multi-step manipulation tasks without robot-specific training
• Demonstrates superior generalization compared to existing methods across diverse scenarios

This research bridges a critical gap in industrial automation by allowing robots to understand and execute complex instructions reliably in real-world settings without extensive retraining for each task.

Data-Agnostic Robotic Long-Horizon Manipulation with Vision-Language-Guided Closed-Loop Feedback