John Phillips*Julieta Martinez*Ioan Andrei Bârsan*Sergio CasasAbbas Sadat and Raquel Urtasun

Companion webpage to our CVPR 2021 publication. Asterisks denote equal contribution.

[Paper PDF] [arXiv] [Video Download] [CVPR '21 Poster]


Over the last few years, we have witnessed tremendous progress on many subtasks of autonomous driving, including perception, motion forecasting, and motion planning. However, these systems often assume that the car is accurately localized against a high-definition map.

In this paper we question this assumption, and investigate the issues that arise in state-of-the-art autonomy stacks under localization error. Based on our observations, we design a system that jointly performs perception, prediction, and localization. Our architecture is able to reuse computation between both tasks, and is thus able to correct localization errors efficiently. We show experiments on a large-scale autonomy dataset, demonstrating the efficiency and accuracy of our proposed approach.

Joint Localization, Perception, and Prediction

The architecture of the combined localization, perception, and prediction (LP2) model.
Proposed Architecture. We propose a joint architecture for localization, perception, and prediction (LP2) from LiDAR data. Please refer to the video above or our paper for additional information.


  title = {Deep Multi-Task Learning for Joint Localization,
           Perception, and Prediction},
  author={Phillips, John and Martinez, Julieta 
          and B{\^a}rsan, Ioan Andrei 
          and Casas, Sergio and Sadat, Abbas 
          and Urtasun, Raquel},
  booktitle = {The {IEEE} Conference on Computer Vision 
               and Pattern Recognition ({CVPR})},
  month = jun,
  year = {2021},
  url = {}