DARPA TIAMAT
Abstract: SRI is developing a neuro-symbolic framework, FLASH: Functionality-based Logic-driven Anchors with Semantic Hierarchy for Fast and Robust Autonomy Transfer, that addresses the Autonomy Transfer Technical Area of the DARPA TIAMAT program. FLASH builds on our neuro-symbolic Trinity architecture developed in the DARPA Assured Autonomy and ARL Internet of Battlefield Things CRA for fast adaptation and robust autonomy in rapidly evolving environments in the presence of shifts in distribution and concept drifts. The transfer of autonomy across domains, where conceptually similar entities can have very different appearances, requires a logical representation of entities that models the affordances and other functionalities. Unlike appearances, the functionalities based on embodied interactions of the agent with the environment are robustly preserved in transfer. Functionality-based representations are naturally hierarchical and, thus, provide efficient and robust conceptual anchors. Further, open-world reasoning is crucial to autonomy transfer to unseen new domains, where not just entities but also behaviors can be new. Hence, we need logical representation that can handle exceptions gracefully and allow non-monotonic reasoning.
Susmit Jha
Technical Director, NuSCI
My research interests include artificial intelligence, formal methods, machine learning and dynamical systems.