ExplainabilitySafetyTransparencycs.CV
Interpretable Underwater Diver Gesture Recognition
Sudeep Mangalvedhekar, Shreyas Nahar, Sudarshan Maskare, Kaushal Mahajan, Anant Bagade
December 8, 2023
arXiv.org
3 citations
Key Takeaways
-
High accuracy alone isn't enough for safety-critical systems: While the model achieves 98.01% accuracy in gesture recognition, the authors recognized that divers need to understand why the system makes decisions underwater—a crucial consideration when miscommunication could pose safety risks.
-
Explainable AI (XAI) techniques make autonomous systems more trustworthy: By applying XAI visualization methods to show which features the model uses to recognize gestures, the researchers demonstrated that transparency in AI decision-making is achievable even in specialized domains like underwater robotics.
-
Responsible AI requires domain-specific considerations: The focus on underwater gesture recognition highlights that responsible AI development must account for unique operational contexts—in this case, the challenges of human-machine communication in extreme environments where divers depend on autonomous vehicles.
-
Interpretability should be built into systems from the start: Rather than treating explainability as an afterthought, the authors integrated XAI techniques directly into their gesture recognition system, suggesting that responsible AI design means planning for transparency during development, not adding it later.
Abstract
In recent years, usage and applications of Autonomous Underwater Vehicles has grown rapidly. Interaction of divers with the AUVs remains an integral part of the usage of AUVs for various applications and makes building robust and efficient underwater gesture recognition systems extremely important. In this paper, we propose an Underwater Gesture Recognition system trained on the Cognitive Autonomous Diving Buddy Underwater gesture dataset using deep learning that achieves 98.01\% accuracy on the dataset, which to the best of our knowledge is the best performance achieved on this dataset at the time of writing this paper. We also improve the Gesture Recognition System Interpretability by using XAI techniques to visualize the model's predictions.