A Deep Learning Method for Vision Based Force Prediction of a Soft Fin Ray Gripper Using Simulation Data.

Area of Science:

• Robotics
• Machine Learning
• Soft Systems

Background:

• Soft robotic grippers are crucial for handling complex, deformable objects.
• Modeling and controlling soft grippers is challenging due to their flexible nature and non-linear dynamics.
• Traditional methods like Finite Element Analysis (FEA) are accurate but computationally expensive for real-time control.

Purpose of the Study:

• To develop a real-time learning-based framework for predicting contact forces and stress distribution in soft robotic grippers.
• To address the data scarcity issue in training such models by leveraging FEA simulations.
• To enable precise grasp planning for delicate objects by providing detailed stress maps.

Main Methods:

• A deep neural encoder-decoder network is proposed to learn internal representations of soft gripper deformations from images.
• The framework predicts contact forces and stress distribution in an end-to-end manner.
• Finite Element Analysis (FEA) is used to generate simulated training data, overcoming limitations of real-world data acquisition.

Main Results:

• The learning-based framework achieves accurate real-time prediction of contact forces and stress distribution.
• FEA-generated data enables better generalizability and faster inference compared to traditional methods.
• The model's predictions are validated against FEA ground truth and real force sensor data under various conditions.

Conclusions:

• The proposed framework offers an efficient and accurate solution for real-time modeling and control of soft robotic grippers.
• Predicting stress distribution provides valuable insights for grasp planning, especially for delicate objects.
• This approach enhances the applicability of soft robotics in complex manipulation tasks.