Event

Paper Release: Semantic Autonomy Stack (SAS) Six-Layer Framework

1 reports · 1 sources

Bogdan Felician Abaza and other authors released a paper proposing the Semantic Autonomy Stack (SAS), a six-layer integrated framework that builds semantic reasoning, task interface, and runtime memory layers on top of validated ROS 2/Nav2, adopting a determinism-first cascading strategy that escalates unresolved or protected cases to a pretrained vision-language model (VLM). The authors state that the frozen v4.8-review implementation was evaluated on two physical robot configurations in an indoor environment, with navigation and perception running onboard and reasoning and memory on a shared workstation; one of the checks measures sampling output stability rather than semantic truth, and one promoted association was reused on the second platform via a deliberately installed shared summary. The authors also state that the larger latency savings are concentrated in paths that bypass VLM inference.