CPU : Central Processing Unit ECU : Electronic Control Unit FPGA : Field Programmable Gate Array
一般道での自動運転実現を目指して, 人工知能（AI)の搭載が可能なAD-ECUを計画中 Aiming to realize automatic driving on general roads, an AD-ECU that can be equipped with Artificial Intelligence is planned.
自動運転ECU ハードウエア AD-ECU Hardware
人工知能(AI)の搭載が可能 Artificial intelligence can be installed.
90%の縮約で約６倍の性能向上 About 6 times performance improvement
by 90% compression rate実装性能評価
Performance Evaluation
DNN : Deep Neural Network
縮約 Compression
学習 Training
実装 Implementation
縮約 Compress
再学習 Re-training
・Acceleratorへ実装 Implemented in ECU accelerator
・枝刈り(識別精度に低寄与な演算の削除) Pruning (Elimination of operations that have low contribution to identification accuracy)
・重み付け Weight Sharing
膨大な情報量 Huge amount of
information
一般道での自動運転に必要なAI(深層学習)を, 低電力・低コストでECUに実装する「AI縮約実装技術」を開発 Hitachi AMS has developed “AI compression technology” that implements AI (Deep Learning) necessary for autonomous driving on general roads in ECUs at low power and low cost.
自動運転ECU ソフトウエアAD-ECU Software
DNN Deep Neural Network
(オリジナル) (original)
DNN Deep Neural Network
(縮約後) (After compression)
性能向上によって「早く, 正確に」周囲の状況を認識できる AD-ECU can recognize the surrounding situation “fast and accurate” by improving performance.
人工知能を搭載すると, 一般道での複雑な走行環境の認識性能が向上 When equipped with artificial intelligence, the recognition performance of complex driving environments on general roads is improved.
人工知能(深層学習)による認識Recognition by Artificial Intelligence (Deep Learning)
車と人が重なっていたり, 一部が隠れている物体も認識 AI can recognize the environment even when a car and a person overlap, or even if an object is partially hidden.
