Electro-Hydraulic Lens Auto Focus Using Reinforcement Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional AF methods for electro-hydraulic (EH) lenses with adjustable focus struggle with dynamic environment changes, moving targets, and designing appropriate reward functions, while also failing to meet real-time focusing requirements.
Innovation Solution
An AF method and system using reinforcement learning (RL) with a deep neural network (DNN) architecture, including state and action spaces, a reward function, and algorithms like DDPG and single hill climbing optimization (HCO), to autonomously adjust focal length and improve focusing accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If reinforcement learning method is used for auto-focusing, then focusing quality and adaptability are improved, but real-time performance and system complexity worsen
Solution Approach 1:
The patent segments the auto-focusing system into distinct functional modules: state space determination module, action space determination module, reward function design module, and policy network execution module. This segmentation allows the complex reinforcement learning system to be managed through modular components, reducing overall system complexity while maintaining adaptability.
Solution Approach 2:
The patent implements preliminary action by pre-defining the state space parameters (image definition, focal length), action space parameters (focusing current value), and reward function structure before actual auto-focusing operations. This preconfiguration reduces real-time computational burden and simplifies the runtime system complexity.
2Reliability
If reinforcement learning method is used for auto-focusing, then focusing quality in dynamic environments is improved, but computational requirements and processing time worsen
Solution Approach 1:
The patent implements feedback mechanisms through the reward function that evaluates image definition changes and provides continuous guidance to the policy network. This feedback loop enables the system to learn optimal focusing strategies over time, improving reliability while the experience pool stores past experiences for faster future decision-making, reducing processing time.
Solution Approach 2:
The patent utilizes parameter changes by dynamically adjusting the state space parameters (image definition, focal length) and action parameters (focusing current) based on real-time feedback. The experience pool stores parameter combinations that have proven effective, allowing the system to quickly retrieve and apply successful parameter settings rather than recomputing them, thus reducing processing time while maintaining reliability.
3Device complexity
If conventional AF method with fixed algorithm is used, then system complexity is reduced, but adaptability to complex environments and moving targets worsens
Solution Approach 1:
The patent applies dynamics by transforming the static fixed algorithm into a dynamic reinforcement learning system. The policy network continuously adapts its focusing strategy based on real-time state observations (image definition, focal length) and receives feedback through the reward function. This dynamic adaptation enables the system to effectively track moving targets and respond to changing environmental conditions while maintaining manageable complexity through modular architecture.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables stable and real-time auto-focusing in dynamic environments, achieving clearer and sharper images by adaptively adjusting focal length based on image quality metrics.
Implementation Method 1
the EH lens with an adjustable focus adjusts a pressure of a liquid by changing an electric field to adjust a focal length
Data Source
AI summary
The present disclosure provides an auto focus (AF) method and system for an electro-hydraulic (EH) lens with an adjustable focus, and an electronic device, and relates to the field of lens focusing technologies. The method mainly includes: determining a state space, an action space, and a reward function of a reinforcement learning (RL) method; constructing sample data based on the foregoing determined state space, action space, and reward function, and storing the sample data in an experience pool; and using sample data in the experience pool as training data of a deep neural network (DNN) architecture when there are M sets of the sample data in the experience pool, and obtaining a trained AF policy with reference to a deep deterministic policy gradient (DDPG) algorithm and a single hill climbing optimization (HCO) algorithm.