[2503.16441] Safe and Efficient Social Navigation through Explainable Safety Regions Based on Topological Features

View the PDF file from the paper entitled safe and effective social navigation through explanatory safety areas on the basis of the Topological features, by Victor Tuscanyo Duran and 4 other authors

PDF HTML (experimental) view

a summary:The last adoption of artificial intelligence in robots is to develop algorithms that enable independent systems to adapt to complex social environments. In particular, safe and effective social navigation is a major challenge, not only Amnesty International requires avoiding deadly collisions and retirement but also to intend intuitively and expected with its surroundings. Methods that depend on the probability models and the generation of the identical safety areas showed promising results in identifying the safety areas with a wrong margin controlled, and primarily dependent on classification approaches and explicit rules to describe the conditions of mobility free of collision. This current perspective work extends through an investigation of how to contribute to the Topological features in creating interpretable safety areas in social mobility scenarios, allowing the classification and description of various simulation behaviors. Instead of relying on behavior parameters to create safety areas, we benefit from topological features through the analysis of the Topological data. We first use the global classification based on the rules to provide interpretable properties of various simulation behaviors, and to distinguish between safe and insecure scenarios based on topological characteristics. Next, we define the safety areas, $ S_ \ Varepsilon $, which represents the areas in the area of ​​the Topological feature where the collisions are avoided with a maximum classification error of $ \ Varepsilon $. These areas are created using adjustive SVM and demand statistics, ensuring strong and developed limits. Our approach initially separates simulation with and without clashes, outperforming methods that do not include topological features. We also improve safety areas to ensure stagnation -free simulations and integrate both sides to determine a compatible simulation space that ensures safe and effective movement.