RoboPilot: Generalizable Dynamic Robotic Manipulation with Dual-thinking Modes

This paper introduces RoboPilot, a dual-thinking closed-loop agentic framework for robotic manipulation that supports adaptive reasoning in dynamic environments. The system dynamically switches between fast and slow thinking to balance efficiency and accuracy. RoboPilot-Bench, a benchmark with 21 tasks across 10 categories, is introduced to evaluate robustness, and experiments show a 11% improvement in task success rate over state-of-the-art baselines.