A Deep Reinforcement Learning Approach to Integrating Vision and Touch in Robotic Arm Control

Reinforcement learning is extensively employed in control tasks of robotic arms. However, factors such as noise and uncertainty introduced by complex environments often lead to the unstable performance of robotic arms. Taking inspiration from babies' tactile exploration during the learning process, we propose a deep reinforcement learning (DRL) method that combines vision and touch to achieve high-performance control of robotic arms based on the deep deterministic policy gradient (DDPG) algorithm. In this approach, the feedback information from the force sensor of the robot arm is fully utilized, and an intrinsic reward mechanism is formulated to encourage interaction between the robot arm and operating objects. This makes up for the deficiency of sparse reward settings that are difficult to provide positive feedback during the early stage of training. Additionally, a contact-prioritized experience replay strategy is adopted to enhance sample utilization, and an asymmetric actor-critic network structure is constructed to overcome challenges associated with directly learning from high-dimensional visual information, thereby improving learning efficiency. Experimental results demonstrate that the proposed algorithm can converge in less time and exhibit superior performance.

• Publication year

  2025

• Venue

  Cybersecurity and Cyberforensics Conference

• Publication date

  2025-07-28

• Fields of study

  Not labeled

• Identifiers
  DOI 10.23919/CCC64809.2025.11179280
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• Exploring Exploration: Comparing Children with RL Agents in Unified Environments

  2020cited by this paper
• Go-Explore: a New Approach for Hard-Exploration Problems

  2019cited by this paper
• Tactile Sensing and Deep Reinforcement Learning for In-Hand Manipulation Tasks

  2019cited by this paper
• Exploration by Random Network Distillation

  2018cited by this paper
• Curiosity-Driven Exploration by Self-Supervised Prediction

  2017cited by this paper
• Hindsight Experience Replay

  2017cited by this paper
• Asymmetric Actor Critic for Image-Based Robot Learning

  2017cited by this paper
• OpenAI Gym

  2016cited by this paper
• Unifying Count-Based Exploration and Intrinsic Motivation

  2016cited by this paper
• Continuous control with deep reinforcement learning

  2015cited by this paper
• Prioritized Experience Replay

  2015cited by this paper
• Active learning of inverse models with intrinsically motivated goal exploration in robots

  2013cited by this paper
• Human Hand Function

  2007cited by this paper
• Human Hand Function

  2006cited by this paper
• Intrinsic and Extrinsic Motivations: Classic Definitions and New Directions.

  2000cited by this paper

• No citing papers are available for this paper.