A Multi-Behavior Planning Framework for Robot Guide
The guiding task of a mobile robot requires not only human-aware navigation, but also appropriate yet timely interaction for active instruction. State-of-the-art tour-guide models limit their socially-aware consideration to adapting to users' motion, ignoring the interactive behavior planning to fulfill the communicative demands. We propose a multi-behavior planning framework based on Monte Carlo Tree Search to better assist users to understand confusing scene contexts, select proper paths and timely arrive at the destination. To provide proactive guidance, we construct a sampling-based probability model of human motion to consider the interrelated effects between robots and humans. We validate our method both in simulation and real-world experiments along with performance comparison with state-of-the-art models.