Talk · International Conference on Information and Communication Technology Convergence · October 20, 2021
A Case Study: Characterization of Performance Inconsistency for Reinforcement Learning on Flappy Bird Game
Aidar Shakerimov, Dmitriy Li, Jurn-Gyu Park
Abstract
One of the serious problems in Reinforcement Learning (RL) algorithms is that their performance usually varies when the same experiment is repeated or reproduced. Although RL results are hard to reproduce due to algorithms' intrinsic variance, which was not investigated systematically. Through this case study on Flappy Bird environment, we introduce and characterize four important factors on performance inconsistency of RL algorithms: 1) level of environment randomness, 2) order of action-value updates process, 3) exploration rate strategy, and 4) selection between on- and off-policy algorithms. Using a quantitative metric (coefficient of variation), we compare, analyze and investigate the results and the effects of each factor on the performance inconsistency/variance in RL. We believe our experimental results and analysis will provide opportunities to obtain an efficient agent that repeats/reproduces more consistent performance results.
More by Aidar Shakerimov
- Design and Control Prismatic Triangular Tensegrity Robot With Linearly Actuated StrutsJournal of Mechanisms and Robotics · 2025 · Citations: 1
- Study on Tensegrity Structure Stability and Payload Manipulability: Triangular vs. Quadrangular DesignsIEEE Workshop on Advanced Robotics and its Social Impacts · 2025 · Citations: 1
- Machine learning-based inverse kinematics scalability for prismatic tensegrity structural manipulatorsRobotics and Autonomous Systems · 2025 · Citations: 9
- QWriter: A Reinforcement Learning-Based Robot for Early Literacy AcquisitionXRDS Crossroads The ACM Magazine for Students · 2023