Hon Hai Research Institute Showcases Breakthrough AI System ModeSeq for Real-Time Traffic Prediction
Hon Hai Research Institute (HHRI), the advanced research and development arm of Hon Hai Technology Group (Foxconn) (TWSE: 2317), has made a significant leap in autonomous driving technology with its state-of-the-art trajectory prediction model, ModeSeq. The breakthrough has positioned HHRI as a global leader in AI innovation, particularly in the field of intelligent vehicle systems.
This milestone achievement was validated on the world stage with ModeSeq taking first place in the prestigious Waymo Open Dataset (WOD) Challenge — specifically in the Interaction Prediction Track — and earning a coveted presentation slot at CVPR 2025, one of the most influential international conferences for computer vision and AI. These honors underscore HHRI’s rising prominence and technical excellence in an increasingly competitive space.
Advancing Safe, Smart Driving Through Multimodal Prediction
“ModeSeq empowers autonomous vehicles with more accurate and diverse predictions of traffic participant behaviors,” explained Dr. Yung-Hui Li, Director of the Artificial Intelligence Research Center at HHRI. “It improves decision-making safety, reduces computational overhead, and uniquely adapts the number of predicted behavior modes based on the uncertainty of each driving scenario.”
Traditional autonomous driving systems often struggle with predicting the multitude of possible movements by surrounding pedestrians and vehicles. ModeSeq addresses this challenge head-on through multimodal trajectory prediction—a technique that considers multiple potential paths, rather than relying on a single anticipated outcome. This approach is critical in high-uncertainty environments, such as intersections or dense urban traffic, where accurate forecasting of human and vehicle movements is essential.
The Technology Behind ModeSeq
Unveiled at CVPR 2025, the research paper titled “ModeSeq: Taming Sparse Multimodal Motion Prediction with Sequential Mode Modeling” was accepted among just 22% of submissions—a testament to its significance in the AI research community.
ModeSeq introduces several innovations:
- Sequential Mode Modeling: This enables the model to represent and rank multiple future paths, dynamically adjusting for uncertainty.
- Early-Match-Take-All (EMTA) Loss Function: A novel training technique that encourages the system to focus on the most plausible future outcomes without sacrificing diversity.
- Advanced Architecture: ModeSeq encodes scene context using Factorized Transformers and decodes it with a hybrid system combining Memory Transformers and custom ModeSeq layers, which enhance performance while maintaining computational efficiency.
The success of ModeSeq doesn’t stop at theory. HHRI’s team further developed Parallel ModeSeq, which claimed first place in the 2025 WOD Challenge at the CVPR WAD Workshop. This real-world validation demonstrated ModeSeq’s ability to outperform entries from global research powerhouses, including the National University of Singapore, University of British Columbia, Vector Institute for AI, University of Waterloo, and Georgia Institute of Technology.
Building on last year’s strong showing—where the original ModeSeq secured second place globally in the 2024 CVPR Waymo Motion Prediction Challenge—this year’s top ranking confirms the system’s growth and technical superiority.
Collaborations Fueling Global Innovation
The research was a collaborative effort led by Director Li and supported by Professor Jianping Wang’s group at the City University of Hong Kong, with additional input from researchers at Carnegie Mellon University. Together, they refined ModeSeq to achieve best-in-class results on the Motion Prediction Benchmark, excelling in metrics like mean Average Precision (mAP) and soft-mAP, while maintaining competitive scores for minimum Average Displacement Error (minADE) and Final Displacement Error (minFDE).
Figure 1: The ModeSeq workflow visualizes potential vehicle paths using red icons and trajectory lines. Each predicted path receives a confidence score, helping the system weigh the likelihood of each scenario.
Figure 2: Director Yung-Hui Li and researcher Ming-Chien Hsu present ModeSeq advancements at CVPR 2025.
Hon Hai Research Institute
Founded in 2020, Hon Hai Research Institute serves as the innovation engine of Foxconn, the world’s largest electronics manufacturer and technology service provider. HHRI comprises five research centers and one advanced laboratory, each focused on long-term innovations within a 3–7 year horizon.
The institute plays a central role in supporting Foxconn’s transformation into a “Smart First” company, aligned with its “3+3+3” strategic model, which emphasizes three emerging industries, three core technologies, and three global markets.
Through pioneering technologies like ModeSeq, HHRI is not only pushing the boundaries of artificial intelligence and autonomous mobility but also reinforcing Foxconn’s broader mission of driving smarter, safer, and more connected futures.