Driving assistance system that realizes anticipatory safety control based on the acquisition of knowledge information for hazard prediction

2021 Fiscal Year Final Research Report

• Project/Area Number: 19K14939
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 20020:Robotics and intelligent system-related
• Research Institution: University of Tsukuba
• Principal Investigator: Saito Yuchi 筑波大学, システム情報系, 助教 (90770470)
• Project Period (FY): 2019-04-01 – 2022-03-31
• Keywords: 安全 / 危険予測 / データ駆動 / 運転支援 / 人間機械系 / ヒューマンインタフェース

Outline of Final Research Achievements

This research project conducted a fundamental study of anticipatory driving support technology with built-in prediction of latent risk based on the context information of the driving environment and the driver behavior status, with the goal of enhancing driving intelligence pertaining to the hazard anticipatory driving in an uncertain environment. We extracted near-miss events from a drive recorder database in situations where accident avoidance is hard for humans to avoid, and successfully classified the data into acceptable and unacceptable outcomes by quantifying the safety margin indices in the near-miss events. This study developed a novel recommended speed AI model driven by data from previous near-miss experiences by learning only acceptable outcomes.

Free Research Field

ヒューマンマシンシステム

Academic Significance and Societal Importance of the Research Achievements

「次に何が起こりうるか」に係る危険予測AIの実現に向けた大きな壁は，「何のデータに学ぶか」である．熟練ドライバは，視覚情報だけでなく，想定事象の知識や過去の経験に基づいて意思決定を実行するが，あらゆる運転データを集めることはデータの収集コストが高い．これに対して，本研究では，18年間に渡って経験豊富なタクシーのヒヤリハットデータを収集しているドライブレコーダDBの活用と機械学習を駆使して，従来モデル駆動で実現できなかった「先読み運転」をデータ駆動AIで挑戦したことに学術的意義があり，交通事故のさらなる削減に寄与しうる技術を開発した点で社会的にも意義がある．