Hinst: Human-Like Interactive Instinct Enables Robots to Robustly Accomplish Universal Tasks

Abstract

Robots have enormous potential to assist humans in daily life. However, current robot development and popularization are impeded by its deficient functionality, lengthy task deployment, and vast experimental training. Poor flexibility, poor practicality, and poor universality in task accomplishment hinder robot applications. Here, a human-like interactive instinct (Hinst) and a Hinst robotic architecture are proposed to enable robots to flexibly and robustly accomplish universal tasks in the real world. The Hinst architecture incorporates multimodal senses, logical decision-making, and initiative task-execution, casting robots with human-like instinctive reaction and intelligence expansion. The robot is empowered with instinctive interaction ability relying on its inherent touch-sense-feedback, knowledge-driven cognition, and adaptive-control, while its high-level intelligence of task-planning and action-skills are nurtured from acquired knowledge-learning or human-teaching. The Hinst greatly enhances the efficiency and the success rate of robotic manipulation, skill learning, and task execution. This anthropomorphic architecture offers a generalizable and universal pathway for general-purpose robots (especially humanoid robots) to robustly accomplish complex housework and industrial skilled work.