Related Experiment Video
Updated: Aug 6, 2026
09:00
Insect-controlled Robot: A Mobile Robot Platform to Evaluate the Odor-tracking Capability of an Insect
Published on: December 19, 2016
From Insect Behavior to Transferable Robot Locomotion: Inferring Embodied Locomotor Principles from Limited Data via
Yuchen Wang1, Thirawat Chuthong2, Mitsuhiro Hayashibe3
1Dept. of Robotics, Graduate School of Engineering, Tohoku University, 6-6-01 Aoba, Aramaki, Aoba-ku,, Sendai, Miyagi, Japan, Sendai, Miyagi, 980-8577, Japan.
Bioinspiration & Biomimetics
|July 24, 2026
Summary
This study introduces a data-driven framework using adversarial inverse reinforcement learning (AIRL) to learn insect locomotion control from biological data. The approach enables adaptive leg coordination for bio-inspired robots, even with limited training data.
Area of Science:
- Robotics
- Computational Neuroscience
- Bio-inspired Engineering
Background:
- Insect locomotion offers insights into adaptive control despite small nervous systems.
- Existing bio-inspired robotics control methods often lack flexibility due to manual tuning and predefined rules.
- Extracting computational principles of insect leg coordination is challenging.
Purpose of the Study:
- To develop a data-driven framework for learning insect locomotion control policies.
- To infer latent reward structures and control strategies from biological walking data.
- To enhance the flexibility and transferability of control strategies in bio-inspired robotics.
Main Methods:
- Utilized adversarial inverse reinforcement learning (AIRL) to learn from stick insect walking data.
- Developed a framework to directly learn continuous locomotion control policies.
- Inferred reward structures and control strategies from behavioral demonstrations.
Main Results:
- The learned policy adapted to different environmental conditions with minimal training data.
- The reward network transferred across different dynamic systems and robot morphologies.
- Achieved faster convergence and more biologically consistent gait coordination than reward shaping methods.
- Demonstrated sim-to-real transferability on a physical bio-inspired robot.
Conclusions:
- The proposed AIRL framework provides a transferable, data-driven method for extracting locomotion control from biological behavior.
- This approach enhances adaptive and flexible leg coordination in bio-inspired robots.
- Offers a promising direction for developing more capable and adaptable robotic systems.
