Evaluation of Natural Human Behavior Using a Motion Prediction Neural Network and its Application

2023 Fiscal Year Final Research Report

• Project/Area Number: 21H03479
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Single-year Grants
• Section: 一般
• Review Section: Basic Section 61020:Human interface and interaction-related
• Research Institution: The University of Tokyo
• Principal Investigator: Makino Yasutoshi 東京大学, 大学院新領域創成科学研究科, 准教授 (00518714)
• Project Period (FY): 2021-04-01 – 2024-03-31
• Keywords: 行動予測 / 追従ロボット / 歩行

Outline of Final Research Achievements

The goal of this research was to take advantage of technology that could predict a person's movements in advance using skeletal information. One of the most important achievements during this research period was the identification of the body parts that are important in predicting a person's gait. We found that 3D information from just three points: the chest and both ankles, can be used to make predictions with the same level of accuracy as whole-body information. We also confirmed that when this information is presented to a person, they can make predictions in a similar way. We also investigated which movements can be predicted and which cannot, and evaluated the time at which the prediction error becomes large. Based on these results, we realized a robot that can follow a person from the front by using the prediction information.

Free Research Field

ヒューマンマシンインタラクション

Academic Significance and Societal Importance of the Research Achievements

人の骨格を利用した予測では，それが可能であることは分かっていたものの，どのような情報が重要か，どのような動作なら予測できるかといった部分は明確になっていませんでした．本研究では歩行について重要なのは高々3点の位置情報であることを明らかにしました．この結果より，歩行予測時に利用可能なセンサの幅が広がります．また，人の集団歩行の解析で利用される上部からの映像でこの3点を検出するのは容易なため，対向者の動作予測を反映した解析なども可能になると期待しています．それ以外にも，予測の誤差が最大となる瞬間とその理由の解明，シンプルな予測モデルとの比較などを通し，動作予測の適用範囲を示すことが出来ました．