Mobile robot path planning using human prediction model based on massive trajectories

We propose global path planning method for mobile robot to avoid human based on massive human movement trajectories. Our method measures trajectories using multiple networked LIDARs in a long term. The captured trajectories are symbolized to grid cell sequences in a time unit. The symbol sequences are modeled by Variable Length Markov Model (VLMM). The method calculates human existence probability at every locations in the future time from the passed trajectory based on the learnt model. Finally, the method plans mobile robot path globally under x-y-t configuration space including human presence probabilities. These mechanisms realize flexible and safety human avoidance of mobile robot. We accumulated human trajectories during approx. 1.5 month and confirmed our method planned path more safety than the method using uniform motion model prediction.