Research

Bird's behavior enrichments using reinforcement learning for ecological environments

• Team  Jin Lee; Youngjun Song; Hongju Yang, Mind Computation Lab

• Advisor Professor Seung Wan Hong, Inha University

• Funding details  This project was supported by the National Research Foundation of Korea (NRF) grants funded by the Ministry of Education (Grant Number NRF- 2021R1A2C1004608).

This project aims to include RL-powered ecological participants in an agent-based simulation. From a master's study, I confirmed that RL behavior models can create serendipitous meetings and meet dynamic needs in the given environment, and found possibilities to embrace a wider range of agent types from humans to microorganisms, insects, and wildlife animals. 

This project aims to promote autonomous behavioral responses of the birds. We develop a reinforcement learning (RL)-powered bird simulation model that learns the behavior policy under architectural variations. The bird simulation stems from a physics-based agent with 15 skeletal joints generating behavioral states, including (1) walking and (2) flying. The RL agent learns to control the joints’ rotation and strength, as well as the lift and thrust forces in the unbuilt, thereby autonomously transiting between walking and flying behaviors, covering grounds and airspaces. This new approach is expected to contribute to the creation of human-animal sharable, safe designs in the fields of architecture and ecology.

The mechanism of bird's behavioral transitions: Creating a reward function for learning to walk and fly