车规级自动驾驶软件开发中预测规划算法的性能瓶颈突破研究

打开文本图片集

中图分类号：U462 收稿日期：2026-01-19 DOI:10.19999/j.cnki.1004-0226.2026.04.010

Research on Breakthrough of Performance Bottleneck of Predictive Planning Algorithm in the Development of Vehicle Grade Autonomous Driving Software

Qian Changgui Pan Asia Technical Automotive Center Co.,Ltd.,Shanghai 20oooo,China

Abstract:Predictive plannngalgorithmis thecoreofautonomous driving decision-making system，butitfacesmultipleconstraintsofreal-t,generalization,dlitiiclevelapplicatio.issudystematicallalystfourjore formancebotlenecksoftealgitcuingteoradictinbtweenomputatioalomplexitydeal-ti,nsuicee generalizatiolitepeabiltetareotitsspsteesuriat egiesarepropoed，amelyghtweightndentcitecturesig，oftwareandrdwarecollboratieotiationeedaptationmechanism，andverfiabiltyenancement.Withtheelpofquantitivegoalsandngieringpracticeverfcation，theinfer ence delay is compressed from 100ms to less than 50ms ，and the long tail scene adaptation ratio is increased from 60% to over 85% ， providing system solutions for large-scale commercial use of autonomous driving.

Key words:Autonomous driving;Prediction planning algorithm;Performance botleneck;Lightweight design

1前言

自动驾驶的决策关键是准确预测周边交通参与者的行为，然后据此规划安全行驶轨迹。（剩余5637字）