From Prediction to Simulation: Unveiling HHRI's Cutting-Edge Breakthroughs in Autonomous Driving Technology

This presentation introduces three significant research achievements from Hon Hai Research Institute in the field of autonomous driving technology: QCNet, BehaviorGPT, and OccNet.

QCNet is an innovative model for trajectory prediction that has won first place in the Argoverse 1 and 2 trajectory prediction challenges for two consecutive years. It employs an efficient Transformer architecture, capable of step-by-step planning and self-correction, significantly improving prediction accuracy and efficiency.

BehaviorGPT is an intelligent simulation agent used for closed-loop testing of autonomous driving algorithms. It achieved the best performance in the 2024 Waymo Sim Agents Challenge, generating near-realistic traffic scenarios that aid in the development and testing of autonomous driving systems.

OccNet is an occupancy network model that constructs high-precision 3D scenes using low-cost equipment. It addresses the issues of 3D point cloud sparsity and multi-modal fusion, outperforming existing state-of-the-art methods in image segmentation tasks.

These research outcomes have not only achieved excellent results in international competitions but have also resulted in multiple top-tier conference papers and patent applications, demonstrating Hon Hai Research Institute's innovative capabilities and leading position in the field of autonomous driving technology.

本次演講我會介紹鴻海研究院在自動駕駛技術領域的三項重要研究成果:QCNet、BehaviorGPT 和 OccNet。

QCNet 是一個用於軌跡預測的創新模型, 在 Argoverse 1 和 2 軌跡預測挑戰賽中連續兩年獲得世界第一。它採用高效的 Transformer 架構, 能夠進行逐步規劃和自我修正, 大幅提升預測準確度和效率。

BehaviorGPT 是一個智能模擬代理器, 用於自駕算法的閉環測試。它在 2024 年 Waymo Sim Agents Challenge 中獲得最佳成績, 能夠生成近真實的交通場景, 有助於自動駕駛系統的開發和測試。

OccNet 是一個佔據網路模型, 能夠利用低成本設備構建高精度的 3D 場景。它解決了 3D 點雲稀疏性和多模態融合的問題, 在影像分割任務中超越了現有的最先進方法。

這些研究成果不僅在國際競賽中取得優異成績, 還產生了多篇頂級會議論文和專利申請, 展現了鴻海研究院在自動駕駛技術領域的創新能力和領先地位。