CVPR Daily - Wednesday

Multi-agent trajectory prediction is essential for autonomous driving to predict what surrounding objects are going to do. Past trajectories will typically be analyzed to forecast where things will move in the future. However, what we know about how objects moved in the past may not be accurate, and if this input is imperfect, how can the prediction of future trajectories be robust? Xinshuo Weng is a PhD candidate in the Robotics Institute at Carnegie Mellon University, working with Professor Kris Kitani, and will shortly be working as a Research Scientist at NVIDIA Autonomous Vehicle Research under the supervision of Marco Pavone. Her paper explores an important and common problem in the autonomous driving domain: multi-agent trajectory prediction. She speaks to us ahead of her poster today. Whose Track Is It Anyway? Improving Robustness to Tracking Errors With Affinity-Based Trajectory Prediction 3 DAILY CVPR Wednesday Xinshuo Weng