Explainable Hybrid CNN and FNN Approach Applied on Robotic Wall-Following Behaviour Learning
- Publisher:
- ACM
- Publication Type:
- Conference Proceeding
- Citation:
- ACM International Conference Proceeding Series, 2021, pp. 623-628
- Issue Date:
- 2021-09-24
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
Fuzzy Neural Network (FNN) applied to robotic control tasks has proved to be effective by previous researchers. However, FNN has an inherent deficiency in dealing with inputs of large dimensions, such as images. Therefore, this research utilizes a Convolutional Neural Network (CNN) model to convert image into distance values and delivers these values to FNN based robot controller as inputs. The proposed hybrid CNN+FNN are tested with both a regression model and a multi-task model. Results show that the multi-task method performs better with less information loss from input images. This paper also proved that the proposed hybrid approach can be generalized into an unknown robotic simulation environment and performs better than its FNN counterpart. By utilizing state of the art explainable analysis method, both the CNN part and the FNN part of the hybrid approach can be explained in a human-understandable way.
Please use this identifier to cite or link to this item: