Workflow
MixedGaussianFlow(MGF)
icon
Search documents
基于扩散模型的多智能体轨迹预测方法1v6小班课来了!
自动驾驶之心· 2025-08-11 05:45
Group 1 - The core focus of the research is on "multi-agent trajectory prediction methods based on diffusion models," which is crucial for applications in autonomous driving, intelligent monitoring, and robot navigation [1][2] - Traditional methods for trajectory prediction often rely on recurrent neural networks, convolutional networks, or graph neural networks, while diffusion models have shown significant improvements in multimodal modeling capabilities [1] - The Leapfrog Diffusion Model (LED) has demonstrated a 19-30 times acceleration in real-time prediction accuracy on datasets such as NBA, NFL, SDD, and ETHUCY [1] Group 2 - The research aims to integrate diffusion generation mechanisms to model trajectory uncertainty while incorporating social interaction modeling and conditional control mechanisms [2] - The expected outcomes include an algorithm framework, quantitative and visual displays, and high-level papers with broad application prospects in autonomous driving, intelligent monitoring, and service robots [2] Group 3 - The course is designed to help students systematically master key theoretical knowledge in trajectory prediction and related fields, addressing gaps in understanding and practical skills [5] - It targets students at various academic levels (bachelor's, master's, PhD) who are interested in trajectory prediction and autonomous driving, aiming to enhance their research capabilities and resume value [7] Group 4 - The course will provide access to public datasets such as ETH, UCY, and SDD, along with baseline code for diffusion model trajectory prediction [19][20] - Students will engage with classic and cutting-edge papers, learning about innovative points, baseline methods, datasets, and writing techniques [5][8]