Multimodal Trajectory Prediction Mannequin?Competitively Acknowledged Internationally
Hon Hai Analysis Institute (HHRI), an R&D powerhouse of Hon Hai Know-how Group (Foxconn) (TWSE: 2317), the world’s largest electronics producer and know-how service supplier, has been acknowledged for its aggressive work in trajectory prediction in autonomous driving know-how.
The landmark achievements in ModeSeq, taking high spot within the Waymo Open Dataset Problem and presenting at CVPR 2025, among the many world’s most influential AI and laptop imaginative and prescient conferences, gathering top-tier tech corporations, analysis establishments, and tutorial leaders, spotlight HHRI’s rising management and technical excellence on the worldwide stage.
“ModeSeq empowers autonomous autos with extra correct and various predictions of visitors participant behaviors,” stated Yung-Hui Li, Director of the Synthetic Intelligence Analysis Middle at HHRI. “It immediately enhances decision-making security, reduces computational value, and introduces distinctive mode-extrapolation capabilities to dynamically alter the variety of predicted habits modes based mostly on state of affairs uncertainty.”
HHRI’s Synthetic Intelligence Analysis Middle, in collaboration with Metropolis College of Hong Kong, on June 13, offered “ModeSeq: Taming Sparse Multimodal Movement Prediction with Sequential Mode Modeling” at CVPR 2025(IEEE/CVF Convention on Pc Imaginative and prescient and Sample Recognition), the place its paper was amongst solely the 22% that have been accepted.
The multimodal trajectory-prediction know-how overcomes the restrictions of prior strategies by each preserving excessive efficiency and delivering various potential final result paths. ModeSeq introduces sequential sample modeling and employs an Early-Match-Take-All (EMTA) loss perform to bolster multimodal predictions. It encodes scenes utilizing Factorized Transformers and decodes them with a hybrid structure combining Reminiscence Transformers and devoted ModeSeq layers.
The analysis crew additional refined it into Parallel ModeSeq, which claimed victory within the prestigious Waymo Open Dataset (WOD) Problem – Interplay Prediction Observe on the CVPR WAD Workshop. The crew’s profitable entry surpassed sturdy rivals from the Nationwide College of Singapore, College of British Columbia, Vector Institute for AI, College of Waterloo and Georgia Institute of Know-how.
Constructing on their success from final 12 months – the place ModeSeq positioned second globally within the 2024 CVPR Waymo Movement Prediction Problem – this 12 months’s Parallel ModeSeq emerged triumphant within the 2025 Interplay Prediction observe.
Led by Director Li of HHRI’s AI Analysis Lab, in collaboration with Professor Jianping Wang’s group at Metropolis College of Hong Kong and researchers from Carnegie Mellon College, ModeSeq outperforms earlier approaches on the Movement Prediction Benchmark—attaining superior mAP and soft-mAP scores whereas sustaining comparable minADE and minFDE metrics.
SOURCE: Foxconn
