Robots operating in unknown environments must gather task-relevant information before acting, but knowing when to stop is an open challenge. A new arXiv preprint proposes a conceptual hybrid neuro-symbolic framework for "task-sufficient active perception," aiming to let a mobile manipulator decide not only how to collect more data but also when further perception is no longer necessary.
After interpreting a requested task, the agent generates an initial plan and iteratively reevaluates it as new information arrives. At each step, it can choose to acquire a more informative viewpoint, inspect or manipulate candidate objects, ask the user for clarification, or perform the task. Neural perception modules update an object-centric belief state, while a symbolic decision layer judges whether the remaining uncertainty justifies additional information gathering.
The framework is still conceptual; the authors plan to implement it on a mobile manipulator with a humanoid torso and mobile base, with complementary evaluation in simulation. The contribution lies in explicitly addressing the stopping problem, which is essential for efficient and timely task execution in real-world settings.