A. Corona, R. Soto, A. Diaz, and J.L. Gordillo (USA)
AGV´s, traversability, active learning, fuzzy logic
Traversability is a measurement which calculates the cost for an autonomous vehicle to move from one point to another related with the condition of the terrain. This cost is normally considered in its path planner algorithm in order to optimize not only the distance traveled by the vehicle but also the condition of the road to be followed. In this work, we propose an ANFIS based architecture to be used as a part of a John Deere´s autonomous vehicle implementation. The architecture also includes the use of an active learning algorithm to define which samples are more valuables to be labeled in order to use them in the learning process. Also, a fuzzy clustering and fuzzy pattern recognition algorithms are compared with ANFIS to define the best value for a certain environment condition to be considered within the path planners of the vehicle. Preliminary results in Table 1 show that predictions of traversavility computation with the proposed architecture are satisfactory for AGV´s applications.
Important Links:
Go Back
