MultiXNet: Multiclass Multistage Multimodal Motion Prediction

• Main Authors: Djuric, Nemanja, Cui, Henggang, Su, Zhaoen, Wu, Shangxuan, Wang, Huahua, Chou, Fang-Chieh, Martin, Luisa San, Feng, Song, Hu, Rui, Xu, Yang, Dayan, Alyssa, Zhang, Sidney, Becker, Brian C, Meyer, Gregory P, Vallespi-Gonzalez, Carlos, Wellington, Carl K
• Format: Journal Article
• Language: English
• Published: 02.06.2020
• Subjects: Computer Science - Computer Vision and Pattern Recognition; Computer Science - Learning
• DOI: 10.48550/arxiv.2006.02000

One of the critical pieces of the self-driving puzzle is understanding the surroundings of a self-driving vehicle (SDV) and predicting how these surroundings will change in the near future. To address this task we propose MultiXNet, an end-to-end approach for detection and motion prediction based directly on lidar sensor data. This approach builds on prior work by handling multiple classes of traffic actors, adding a jointly trained second-stage trajectory refinement step, and producing a multimodal probability distribution over future actor motion that includes both multiple discrete traffic behaviors and calibrated continuous position uncertainties. The method was evaluated on large-scale, real-world data collected by a fleet of SDVs in several cities, with the results indicating that it outperforms existing state-of-the-art approaches.