SOURCE-LINKED INTELLIGENCE

Social Navigation with Adaptive Behavior Sets

CORDIS · observation · Publication date unknown

ch as between areas where more safety space should be given than in others (e.g. hospitals compared to offices). For this purpose, SNABS (Social Navigation with Adaptive Behavior Sets) investigates a machine learning framework (Adaptive Behavior Sets for Transfer Reinforcement Learning) that promises quick adaptations by learning an extensive library of potential behaviors. Being in a new context, it selects the best behavior among them. However, the existing framework suffers from problems hindering its use for real-world applications. This includes the prediction of its behaviors’ outcomes in complex navigation scenarios with human trajectories and the identification of the best behavior in real-time. SNABS aims to overcome these problems by investigating new generative AI models for predicting future outcomes and by introducing new selection strategies for the behaviors. The project further researches how to successfully use the framework to train an adaptive social navigation controller for social robots. The controller will be evaluated in realistic healthcare scenarios. During

Read original source ↗ Open in workspace

recordType

award

status

SIGNED

region

EU

value

390916.86

unit

EUR