Machine learning-based framework for wall-perching prediction of flying robot.

Area of Science:

• Robotics
• Machine Learning
• Bio-inspired Engineering

Background:

• Animals exhibit diverse natural perching behaviors, but human-made aircraft struggle with vertical wall landings.
• Traditional dynamic simulations and experiments are inefficient for analyzing and designing perching robots.
• Developing efficient methods for vertical-wall perching is crucial for advancing flying robot capabilities.

Purpose of the Study:

• To develop an efficient machine learning framework for predicting vertical-wall perching success in flying robots.
• To overcome the inefficiencies of traditional dynamic simulation and experimental approaches.
• To optimize robot control and structural parameters for stable perching.

Main Methods:

• A validated knowledge-based model was used to compute transient dynamics during high-speed perching.
• Key success factors for perching were identified using the knowledge-based model.
• A data-driven machine learning model was trained on mixed sample data to predict perching success or failure.

Main Results:

• The machine learning framework accurately predicts the success or failure of arbitrary perching events.
• High-precision predictions enable optimization of robot control and structural parameters.
• The developed framework significantly reduces the time and cost associated with conventional design approaches.

Conclusions:

• The proposed machine learning framework offers an efficient and high-precision solution for vertical-wall perching prediction.
• This advancement optimizes flying robot design and control, enhancing their capabilities for complex landing scenarios.
• The study paves the way for faster, more cost-effective development of perching robots.