ML-based Motion Prediction

ktro2828 commented 8 months ago

Checklist

[X] I've read the contribution guidelines.
[X] I've searched other issues and no duplicate issues were found.
[X] I've agreed with the maintainers that I can plan this task.

Description

Integrate ML-based motion prediction to the Autoware perception stack to improve the reliability and accuracy of the predicted paths.

Current rule-based prediction can only estimate paths aligned to the center line of the road for vehicles or straight paths for other traffic participants (e.g. pedestrian, cyclists).

ML-based module allow us to predict their future behaviors by considering observed state and their interactions.

Purpose

To improve accuracy, adaptability and be able to handle uncertanity of future behaviors of traffic participants.

Possible approaches

Now we are trying to integrate MTR(Motion TRnsformer), which achieved the 1st place for the motion prediction task in 2022 Waymo Open Dataset Challege.

Paper: https://arxiv.org/abs/2209.13508
Original repo: https://github.com/sshaoshuai/MTR

Definition of done

Implementation to Autoware

sezan92 commented 8 months ago

is it possible to work on this ?

ktro2828 commented 8 months ago

@sezan92 Thanks for your interest. I've finished creating the training pipeline of MTR. Now I'm working on the task to support inference with TensorRT, and integrate to Autoware as ROS 2 node.

sezan92 commented 8 months ago

@sezan92 Thanks for your interest. I've finished creating the training pipeline of MTR. Now I'm working on the task to support inference with TensorRT, and integrate to Autoware as ROS 2 node.

damn, missed it! :(

zehranrgi commented 6 months ago

Hi, what were your steps to integrate another model into Autoware?

stale[bot] commented 4 months ago

This pull request has been automatically marked as stale because it has not had recent activity.

autowarefoundation / autoware.universe