Research Analyzer
← Back ICRA 2026

A Policy Model Based Efficient and Accurate Scene Recognition Method for Service Robot

Shaopeng Liu, Guanzhong Zhou, Chao Huang

PDF

AI summary

Key figure (auto-extracted from paper)
A unified DQN-based policy model jointly evaluates image similarity and adjusts camera viewpoints to achieve superior accuracy and efficiency in robotic scene recognition.
Scene recognition topological semantic map active vision deep Q-learning service robots policy model

Problem

Current scene recognition methods for service robots suffer from low accuracy due to poor image quality and low efficiency from redundant processing of similar adjacent images.

Approach

The method uses a deep Q-learning network to simultaneously decide whether to inherit a scene label from the previous node or adjust the robot's viewing angle, guided by a custom reward function and generative path training.

Key results

  • Unified DQN policy model for joint similarity judgment and view adjustment
  • Rule-based reward function with scene score model for effective policy co-learning
  • Generative path-based training strategy to synthesize sufficient training data
  • State-of-the-art accuracy and efficiency in simulation and real-world mobile robot deployment

Why it matters

Enables service robots to construct accurate topological semantic maps in real-time with reduced computational overhead, advancing practical domestic automation.

Abstract

In domestic environments, assigning scene semantic labels (scene recognition) to each node of a topological semantic map is an important task. Given the limitations of current scene recognition methods in efficiency, and accuracy for service robot, this letter proposes a scene recognition method based on a policy model. Considering the similarity of images captured from the adjacent nodes and the low-quality image caused by the uncertain node position and observation direction of the robot, we develop a policy model using a deep Q-learning network (DQN). This model enhances accuracy and efficiency by deciding whether to (1) inherit thescenetypefromtheprecedingnodewithoutre-recognitionor(2) adjust the robot’s observation angle to capture a more informative image. A rule-based reward function integrated with a scene score model enables simultaneous learning of similarity assessment and viewpoint adjustment policies. Furthermore, a training strategy based on generated path is proposed to provide sufficient data for training the policy model. Extensive comparative experiments in simulated environments demonstrate that our method surpasses state-of-the-art approaches in both recognition accuracy and effi- ciency.Deploymentonamobilerobotconfirmsitspracticalefficacy, achieving precise and efficient scene recognition across diverse real-world environments.

Index terms

Recognition Cognitive Modeling Reinforcement Learning

Related papers